JP2012113625A - Information processing apparatus, information processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To achieve fast clustering processing which suppresses throughput.SOLUTION: An information processing apparatus 100 includes: an N-ary numeral generation unit 101 which generates an N-ary numeral in which values of coordinates of respective dimensions represented in N-ary (N=2, 3, ...) representation of a predetermined number of digits are arrayed by one digit for each dimension in order with respect to data 1011 having position information on a feature space 1001 defined with D-dimensional coordinates (D=2, 3, ...); and a clustering unit 103 which classifies the data 1011 in which high-order (k) digits (k=1, 2, ...) of the N-ary numeral are common into the same cluster 1021.

Description

  The present invention relates to an information processing apparatus, an information processing method, and a program.

  A technique for clustering data on a feature space based on position information included in the data is known. Data classified into the same cluster by clustering is considered to be data at a close position in the feature space, that is, data having similar features represented by the feature space. As a technique using such clustering, for example, Patent Document 1 describes a technique of adding position information of shooting locations to image data and classifying the image data for each shooting location by clustering based on the position information. Has been.

JP 2010-140383 A

  However, since the clustering process calculates a distance between each of a plurality of pieces of data having position information, the processing load of the distance calculation tends to increase. Also, it tends to require a large amount of memory. Therefore, improvement of the processing speed of the clustering process has been a problem.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a new and improved information processing apparatus, information processing method, and information processing apparatus capable of high-speed clustering processing with a reduced processing amount. To provide a program.

  In order to solve the above-described problem, according to one aspect of the present invention, a predetermined number of digits is obtained for data having position information of a feature space defined by D-dimensional coordinates (D = 2, 3,. An N-ary value generation unit that generates an N-ary value in which the values of coordinates of each dimension expressed in N-ary numbers (N = 2, 3,. There is provided an information processing apparatus including a clustering unit that classifies the above-mentioned data having the same high-order k digits (k = 1, 2,...) Into the same cluster.

The clustering unit, k = D × m (m = 1,2, ···) when it is, the data higher k digits of the N-ary number is common, m layer of ^{N D-ary} tree structure Cluster You may classify into the same cluster visually.

  The clustering unit may include a clustering sorting unit that sorts the data in the order of the N-ary value, and may specify the data classified into the same cluster from a result of the sorting.

  The clustering unit is a cluster that identifies the one cluster based on a first position where the data classified into one cluster appears in the sorting result and the number of the data classified into the one cluster. Specific information may be generated.

  The information processing apparatus includes: a merge sorting unit that sorts the clusters with respect to a first direction in the feature space based on a result of a first rank determination process based on the D-dimensional coordinates; and the first direction. An adjacency determination unit that determines whether or not the clusters sorted for are adjacent to each other in the first direction, and further includes a merge unit that merges the clusters determined to be adjacent to each other in the first direction You may prepare.

  The merge sorting unit sorts the clusters in the second direction in the feature space based on the result of the second rank determination process based on the D-dimensional coordinates, and the adjacency determination unit The clusters sorted in the second direction are determined to be adjacent to each other in the second direction, and the merge unit may further merge the clusters determined to be adjacent to each other in the second direction. Good.

  The feature space is a ground surface, the D-dimensional coordinates are two-dimensional coordinates composed of latitude and longitude, and the cluster is included in a grid defined on the ground surface by the two-dimensional coordinates. The first rank determination process sorts the grid in the first direction and gives the sorting order of the grid as a rank to the clusters included in each grid. It may be a process.

  The feature space is a three-dimensional space, the D-dimensional coordinates are three-dimensional coordinates constituting an orthogonal coordinate system, and the cluster is defined in the three-dimensional space by the three-dimensional coordinates. It may be an area including the position information of the data included in the block.

  In order to solve the above problem, according to another aspect of the present invention, data having position information in a feature space defined by D-dimensional coordinates (D = 2, 3,...) Generating an N-ary value in which the coordinate values of each dimension expressed in N-digits (N = 2, 3,. And classifying the data having the same upper k digits (k = 1, 2,...) Into the same cluster.

  In order to solve the above problem, according to still another aspect of the present invention, data having position information in a feature space defined by D-dimensional coordinates (D = 2, 3,...) A process of generating an N-ary value in which the coordinates of each dimension expressed in N-digit numbers (N = 2, 3,...) Having a predetermined number of digits are arranged in order for each dimension, and N There is provided a program for causing a computer to execute processing for classifying the above-mentioned data having the same high-order k digits (k = 1, 2,...) Into the same cluster.

  The program includes a process of sorting the clusters with respect to a first direction in the feature space based on a result of a first rank determination process based on the D-dimensional coordinates, and a cluster sorted with respect to the first direction. However, the computer may further execute a process of determining whether or not adjacent to each other in the first direction and a process of merging clusters determined to be adjacent to each other in the first direction.

  The process of merging the clusters may include a process of calculating a distance between the clusters and a process of merging the clusters when the calculated distance is equal to or less than a predetermined threshold.

  The process of merging the clusters includes a process of calculating a distance between the clusters, a process of storing the cluster as a merge candidate cluster when the calculated distance is equal to or less than a predetermined threshold, and the storage The merge candidate clusters may be merged in order from the merge candidate clusters with the smallest distance between them.

  As described above, according to the present invention, high-speed clustering processing with a reduced processing amount is possible.

It is a figure which shows the example of the relationship of the content in the 1st Embodiment of this invention, a cluster, and a grid. It is a figure which shows the example of the hierarchical structure of the grid in the 1st Embodiment of this invention. It is a figure which shows the example of the result of the clustering in the 1st Embodiment of this invention. It is a figure for comparing and comparing the grid-based clustering and the general distance-based clustering in the first embodiment of the present invention. It is a figure for comparing and comparing the grid-based clustering and the general distance-based clustering in the first embodiment of the present invention. It is a block diagram which shows the structure of the information processing apparatus which concerns on the 1st Embodiment of this invention. It is a figure for demonstrating the clustering in the 1st Embodiment of this invention. It is a figure for demonstrating the merge of the cluster in the 1st Embodiment of this invention. It is a flowchart which shows the clustering and merge process in the 1st Embodiment of this invention. It is a figure for demonstrating the clustering in the 1st Embodiment of this invention. It is a figure for demonstrating the cluster specific information in the 1st Embodiment of this invention. It is a flowchart which shows the merge related process in the 1st Embodiment of this invention. It is a figure which shows the example of the merge setting information in the 1st Embodiment of this invention. It is a flowchart which shows the merge setting selection process in the 1st Embodiment of this invention. It is a flowchart which shows the search order merge process in the 1st Embodiment of this invention. It is a flowchart which shows the full match merge process in the 1st Embodiment of this invention. It is a figure which shows the grid search in the horizontal direction in the 1st Embodiment of this invention. It is a figure which shows the grid search in the perpendicular direction in the 1st Embodiment of this invention. It is a figure which shows the grid search in the diagonally lower right direction in the 1st Embodiment of this invention. It is a figure which shows the grid search in the diagonal upper right direction in the 1st Embodiment of this invention. It is a figure for demonstrating the one way search in the 1st Embodiment of this invention. It is a figure for demonstrating the two-way search in the 1st Embodiment of this invention. It is a figure for demonstrating the four-way search in the 1st Embodiment of this invention. It is a flowchart which shows the vicinity search merge process (no upper search) in the 1st Embodiment of this invention. It is a flowchart which shows the adjacent search process (no high-order search) in the 1st Embodiment of this invention. It is a flowchart which shows the neighborhood search merge process (with high rank search) in the 1st Embodiment of this invention. It is a figure which shows the example of the high-order grid list | wrist in the 1st Embodiment of this invention. It is a figure which shows the example of the high-order grid list | wrist in the 1st Embodiment of this invention. It is a figure for demonstrating the adjacent search process (with high rank search) in the 1st Embodiment of this invention. It is a flowchart which shows the adjacent search process (with high-order search) in the 1st Embodiment of this invention. It is a figure for demonstrating the adjacent search process (with high rank search) in the 1st Embodiment of this invention. It is a figure which shows the example of the merging object grid of the vicinity search merge process (with high-order search) in the 1st Embodiment of this invention. It is a figure for demonstrating the outline | summary of the distance order sort in the 1st Embodiment of this invention. It is a flowchart which shows the distance order merge process in the 1st Embodiment of this invention. It is a figure which shows the example of the relationship of the content in the 2nd Embodiment of this invention, a cluster, and a block. It is a figure which shows the example of a display of the content and cluster in the 2nd Embodiment of this invention. It is a figure for demonstrating the division | segmentation of the ground surface by the block in the 2nd Embodiment of this invention. It is a figure for demonstrating the division | segmentation of the ground surface by the block in the 2nd Embodiment of this invention. It is a figure for demonstrating the division | segmentation of the ground surface by the block in the 2nd Embodiment of this invention. It is a figure for demonstrating the clustering in the 2nd Embodiment of this invention. It is a block diagram for demonstrating the hardware constitutions of the information processing apparatus which concerns on embodiment of this invention.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In the present specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant description is omitted.

The description will be made in the following order.
1. 1. First embodiment 1-1. Outline of grid-based position clustering 1-2. Configuration of information processing apparatus 1-3. 1. Details of clustering and merge processing Second embodiment 2-1. 2. Overview of block-based location clustering 3. Hardware configuration of information processing apparatus according to embodiment of the present invention Summary

<1. First Embodiment>
In the first embodiment of the present invention, the ground surface corresponds to the feature space. In the present embodiment, position information on the ground surface is represented by two-dimensional coordinates such as latitude and longitude. In the present embodiment, the cluster is an area that includes position information of content included in a grid defined on the ground surface by latitude and longitude.

(1-1. Overview of grid-based location clustering)
First, an overview of clustering in the first embodiment of the present invention will be described with reference to FIGS. Clustering in the present embodiment classifies content having position information into clusters with reference to a grid defined by two-dimensional coordinates of latitude and longitude, and can be said to be grid-based position clustering.

(About the grid)
FIG. 1 is a diagram illustrating an example of the relationship among content, clusters, and grids according to the first embodiment of the present invention. In FIG. 1, the ground surface 1001, the content 1011, the cluster 1021, and the grid 1031 are illustrated.

  The ground surface 1001 is a region of all or part of the surface of the earth. In the present embodiment, the ground surface 1001 is treated as a two-dimensional plane in which position information is represented by two-dimensional coordinates such as latitude and longitude.

  The content 1011 is data having position information for specifying a position on the ground surface 1001. The content 1011 may be position information itself, or may be data to which position information is added as additional information for some information. The content 1011 can be, for example, image data to which position information on the shooting location is added.

  The cluster 1021 is an area including the content 1011 located at a position close to each other on the ground surface 1001. The cluster 1021 is illustrated as a substantially rectangular shape, but may have other shapes. The cluster 1021 may be a circumscribed figure of the content 1011 included in the cluster 1021.

  The grid 1031 is a grid set on the ground surface 1001. The grid 1031 can be a rectangular area defined by latitude and longitude ranges on the ground surface 1001. As will be described later, the size of the grid 1031 can be set to an appropriate size according to the clustering conditions such as the number of contents 1011 and the size of a region to be clustered.

  As illustrated, in the present embodiment, the content 1011 included in the same grid 1031 is classified into the same cluster 1021. The area of one cluster 1021 into which the content 1011 is classified is included in the area of one grid 1031 except when the cluster 1021 is merged. That is, in the grid-based position clustering process in the present embodiment, whether or not they are included in the same grid 1031 is a basic standard for clustering.

  Here, in general distance-based position clustering, a distance between contents is calculated, and a process of comparing the distance with a predetermined threshold or a distance between other contents is performed. In such processing, the distance calculation is performed by the number of combinations of contents, so the processing load of distance calculation increases. Further, when comparing the distance between contents with the distance between other contents, many storage areas are used to temporarily hold the calculated distance.

  On the other hand, in the grid-based position clustering in this embodiment, the position information itself of the content 1011 represents the grid 1031 including the content 1011. As will be described later, when the content 1011 is sorted in the order of an N-ary value in which latitude and longitude values are arranged one by one in order, the content 1011 included in the region of the same grid 1031 is adjacent to each other in the sorting result. Whether or not the adjacent contents 1011 are included in the same grid 1031 in the sorting result is, for example, whether the upper k digits (k = 1, 2,...) Of the N-ary value are common. Can be determined. Here, as described above, the content 1011 included in the same grid 1031 is classified into the same cluster 1021. Therefore, in the grid-based position clustering in the present embodiment, sorting the numerical values generated from the position information of the content 1011 is the main processing of clustering. The sort process requires a smaller processing load than the above distance calculation and requires fewer processing times. Therefore, in the grid-based position clustering in this embodiment, the processing speed is improved and the storage area to be used is also reduced.

(About the grid hierarchy)
FIG. 2 is a diagram illustrating an example of a hierarchical structure of the grid in the first embodiment of the present invention. FIG. 2 shows a level 0 grid 1032, a level 1 grid 1033, and a level 2 grid 1034.

  The level 0 grid 1032 is a topmost grid in a hierarchical structure that covers the entire ground surface 1001. That is, at the top of the hierarchical structure, the entire ground surface 1001 is included in one grid.

  The level 1 grid 1033 is a grid obtained by dividing the level 0 grid 1032 into two in the latitude direction and the longitude direction. In other words, the level 1 grid 1033 divides the entire ground surface 1001 that is the area of the level 0 grid 1032 into four.

  The level 2 grid 1034 is a grid obtained by dividing the level 1 grid 1033 into two in the latitude direction and the longitude direction. In other words, the level 2 grid 1034 divides the area of the level 1 grid 1033 into four. That is, the level 2 grid 1034 divides the entire ground surface 1001 that is the area of the level 0 grid 1032 into 16 parts.

  The hierarchical structure of the grid is similarly extended to lower levels. Specifically, a finer area grid can be defined such as a level 3 grid obtained by dividing the area of the level 2 grid 1034 into four, a level 4 grid obtained by dividing the area of the level 3 grid into four, and so on. By adjusting the level of the grid used for the clustering process, it is possible to balance the clustering granularity and the processing load.

  Thus, in the present embodiment, a grid obtained by dividing a grid at a certain level into two in the latitude direction and the longitude direction respectively becomes a grid of one level below. In other words, the lower level grid divides the area of the upper level grid into four. Therefore, the hierarchical structure of the grid 1031 in this embodiment has a quadtree structure with the level 0 grid 1032 as a root node and the following grids at each level as nodes, and the cluster 1021 defined according to the grid 1031 also has Have a similar quad-tree structure.

  Here, in general distance-based position clustering, when a tree structure of a cluster is defined, a storage area that holds information on the tree structure is consumed. On the other hand, in the grid-based position clustering according to the present embodiment, since the quadtree structure of the grid 1031 is uniquely determined as described above, if the information about which level the grid 1031 is is retained. Based on the quadtree structure of the grid 1031, it is possible to easily grasp the tree structure of the cluster 1021.

(About clustering results)
FIG. 3 is a diagram illustrating an example of the result of clustering in the first exemplary embodiment of the present invention. FIG. 3 shows a map 1002, a content icon 1012, a cluster area 1022, a cluster center 1023, and a grid line 1035.

  A map 1002 is an image representing a part or all of the area of the ground surface 1001. A map 1002 is displayed to indicate to the user the location of the content 1011 and the area of the cluster 1021 that is the result of clustering the content 1011. The area of the ground surface 1001 represented by the map 1002 may be set according to a range where the content 1011 exists on the ground surface 1001 or may be set according to a user operation.

  The content icon 1012 is displayed at a position on the map 1002 corresponding to the position of the content 1011 on the ground surface 1001. The content icon 1012 is illustrated as a pin-shaped icon, but is not limited to this, and may be an icon of various shapes. The content icon 1012 may display a part or all of information such as characters or images included in the content 1011.

  The cluster area 1022 is displayed at a position on the map 1002 corresponding to the area of the cluster 1021 on the ground surface 1001. The cluster area 1022 may be displayed in the same shape as the area of the cluster 1021. Also, for example, in order to avoid overlapping the display with the content icon 1012 and make the content icon 1012 easier to see, the cluster area 1022 may be displayed. It may be slightly expanded.

  The cluster center 1023 is displayed at a position on the map 1002 corresponding to the center of the cluster region 1022 or the position of the center of the cluster 1021 on the ground surface 1001. The cluster center 1023 may be displayed, for example, to show the user summary information extracted from the content 1011 included in the cluster 1021 (such as a representative image or a thumbnail image if the content 1011 is image data). , Not necessarily displayed.

  The grid line 1035 is a line indicating the grid 1031 used in the clustering for classifying the content 1011 into the cluster 1021. The grid line 1035 is not displayed on the map 1002 which is the display of the clustering result. However, for example, when the user changes the setting of clustering granularity, the grid line 1035 may be displayed as reference information.

(Comparison with distance-based clustering-grid-based advantages)
FIG. 4 is a diagram for comparing and explaining grid-based clustering according to the first embodiment of the present invention and general distance-based clustering. FIG. 4 illustrates a case where the contents 1011a to 1011k are clustered on a distance basis (a) and a case where clustering is performed on a grid base in this embodiment (b).

  As shown in (a), when the contents 1011a to 1011k are clustered on a distance basis, the contents 1011a to 1011e are classified into the cluster 1021a, the contents 1011f to 1011j are classified into the cluster 1021b, and the content 1011k is classified into the cluster 1021c. Is done. As shown in the figure, when the areas of the clusters 1021a to 1021c are illustrated in an oval shape, the area of the cluster 1021a and the area of the cluster 1021b partially overlap each other, and the area of the cluster 1021c is the area of the cluster 1021b. Is included. In the case of distance-based clustering, for example, depending on the distance calculation and comparison procedure, a cluster structure of such a complicated region can be generated.

  On the other hand, as shown in (b), when the contents 1011a to 1011k are clustered on a grid basis, the contents 1011a and 1011b are classified into the cluster 1021d, the contents 1011c are classified into the cluster 1021e, and the contents 1011d to 1011g are classified into the cluster 1021f. The contents 1011h to 1011k are classified into the cluster 1021g. As shown in the figure, the clusters 1021d to 1021g are clearly separated without interfering with each other. As described above, in the case of grid-based position clustering, whether or not the content 1011 is included in the same grid 1031 is a basic criterion for clustering. Therefore, the region of the cluster 1021 is included in the region of the grid 1031 in principle. . Therefore, a cluster structure of intricate regions will not be formed.

(Comparison with distance-based clustering-grid-based weakness)
FIG. 5 is a diagram illustrating another example for comparing and explaining clustering according to the first embodiment of the present invention and general distance-based clustering. FIG. 5 illustrates a case where the contents 1011m to 1011q are clustered on a distance basis (a) and a case where the contents 1011m to 1011q are clustered on a grid basis in this embodiment (b).

  As shown in (a), when the contents 1011m to 1011q are clustered on a distance basis, the contents 1011m and 1011n are classified into the cluster 1021h, and the contents 1011o to 1011q are classified into the cluster 1021i. In the case of distance-based clustering, basically, the content 1011 having a close distance is classified into the same cluster 1021 in this way.

  On the other hand, as shown in (b), when the contents 1011m to 1011q are clustered on a grid basis, the contents 1011m and 1011n are classified into the cluster 1021j, the contents 1011o are classified into the cluster 1021k, and the contents 1011p and 1011q are classified into the cluster 1021m. are categorized. As described above, in the case of grid-based position clustering, whether or not the content 1011 is included in the same grid 1031 is a basic criterion for clustering. Therefore, the content 1011 included in the different grid 1031 Even if the distance is short, it may not be classified into the same cluster 1021.

  Further, in (b), a grid boundary 1036 and an upper grid boundary 1037 are shown. The grid boundary 1036 is a boundary of the grid 1031. The upper grid boundary 1037 is a boundary of the grid 1031 and a boundary of an upper level grid including the four grids 1031. In the illustrated example, the contents 1011m and 1011n are included in the grid 1031a, the content 1011o is included in the grid 1031e, and the contents 1011p and 1011q are included in the grid 1031c.

  Here, consider a case where clustering is performed on the upper grid of the grid 1031. As illustrated, the grid 1031a and the grid 1031e are included in the same upper grid. Therefore, when clustering in the upper grid is executed, the contents 1011m, 1011n, and 1011o are classified into the same cluster 1021n. On the other hand, the grid 1031c is included in a higher-order grid different from the grids 1031a and 1031e. For this reason, even if clustering in the upper grid is executed, the clusters of the contents 1011p and 1011q do not change in the cluster 1021m.

  As described above, the grid-based position clustering has a great advantage that the processing is very fast. However, when the grid 1031 has a boundary, the content 1011 having a short distance is classified into the same cluster 1021. Note that it may not be possible. In such a case, the result of clustering can be approximated to a natural form by a merge process described later. As will be described later, this merging process can also be realized by a high-speed process utilizing the characteristics of grid-based clustering.

(1-2. Configuration of information processing apparatus)
Next, the configuration of the information processing apparatus according to the first embodiment of the present invention will be described with reference to FIGS.

  FIG. 6 is a block diagram showing the configuration of the information processing apparatus according to the first embodiment of the present invention. In FIG. 6, the information processing apparatus 100 and an N-ary value generation unit 101, a clustering unit 103, a merge unit 107, an input unit 113, a display control unit 115, and a display unit 117 that are mainly included in the information processing apparatus 100 are illustrated. ing.

  The information processing apparatus 100 handles the above-described content 1011 as data. The content 1011 can be, for example, image content such as still image content or moving image content, or various character information or image information registered in a server or the like so that users can share various information. The content 1011 may be content such as mail, music, schedule, electronic money usage history, call history, content viewing history, sightseeing information and area information, news and weather forecast, and ringtone mode history. In the following description, image content such as still image content or moving image content will be described as an example. However, the information processing apparatus 100 can handle arbitrary information and content data as long as the position information representing the position in the feature space is data attached as, for example, metadata.

  Moreover, it is preferable that the above-described content data and data representing various types of information are stored inside the information processing apparatus 100. However, the data body may be stored in a device such as a server provided outside the information processing apparatus 100, and the metadata corresponding to these data bodies may be stored in the information processing apparatus 100. Hereinafter, a case where the information processing apparatus 100 stores content data and data representing various types of information together with metadata will be described as an example.

(N-ary value generator)
The N-ary value generation unit 101 is realized by, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and the like. The N-ary numerical value generation unit 101 calculates the latitude and longitude values expressed in binary numbers of a predetermined number of digits for the content 1011 having the position information of the ground surface 1001 defined by the two-dimensional coordinates of latitude and longitude. A binary value is generated in which one digit is arranged in order for latitude and longitude.

For example, when 29 digits are set as the predetermined number of digits, the N-ary value generation unit 101 expresses the values of latitude and longitude as 29-digit binary values, respectively. Here, if the latitude expressed in binary number is “a ₂₈ a ₂₇ a ₂₆ ... A ₀ ” and the longitude expressed in binary number is “b ₂₈ b ₂₇ b ₂₆ ... B ₀ ”, A binary value in which these values are arranged one digit at a time in each dimension is a 58-digit binary value “b ₂₈ a ₂₈ b ₂₇ a ₂₇ b ₂₆ a ₂₆ ... B ₀ a ₀ ”. When the predetermined number of digits is 29, the minimum latitude resolution is equivalent to 20000 km / 2 ²⁹ = 0.04 m when the radius of the earth is 20000 km. In this case, the minimum resolution of longitude is equivalent to 40000 km / 2 ²⁹ = 0.07 m when the diameter of the earth is 40000 km. The predetermined number of digits can be set to an appropriate value in consideration of, for example, the required minimum resolution and the size of the data unit handled by the information processing apparatus 100 (for example, 32 bits, 64 bits, etc.).

  As described above, the N-ary value generation unit 101 generates one binary value from the position and longitude values, thereby making it possible to hold position information that is two-dimensional coordinates as one value. Further, since the binary value of latitude and the binary value of longitude are sequentially arranged in the binary value, the clustering process is facilitated as will be described later.

(Clustering part)
The clustering unit 103 is realized by a CPU, a ROM, a RAM, and the like. The clustering unit 103 includes a clustering sorting unit 105 described later. The clustering unit 103 classifies the content 1011 having the same upper k digits (k = 1, 2,...) Of the binary values generated by the N-ary value generation unit 101 into the same cluster. When k = 2 × m (m = 1, 2,...), the cluster into which the content 1011 having the same upper k digits of the binary value is classified is m layers of 2 ² = quadrant tree cluster. Eyes.

  The clustering sort unit 105 is included in the clustering unit 103 and is realized by a CPU, a ROM, a RAM, and the like. The clustering sorting unit 105 sorts the content 1011 in the order of the binary values generated by the N-ary value generation unit 101. As will be described later, the clustering unit 103 identifies the content 1011 classified into the same cluster from the result of sorting by the clustering sorting unit 105.

  Here, the function of the clustering unit 103 will be described with reference to FIG. FIG. 7 is a diagram for explaining clustering in the first embodiment of the present invention. In FIG. 7, a grid 1031 defined on the ground surface 1001, an upper grid 1041 which is a higher level grid of the grid 1031, and contents 1011 u to 1011 w to be clustered are illustrated.

  As described with reference to FIG. 1, in this embodiment, the cluster 1021 into which the content 1011 is classified is defined by the grid 1031 to which the content 1011 belongs. That is, in the clustering stage, if the grid 1031 to which the content 1011 belongs is specified, the grid 1031 to which the content 1011 is automatically classified is specified. Therefore, classifying the content 1011 into the cluster 1021 by the clustering unit 103 is substantially the same as determining the grid 1031 to which the content 1011 belongs. Therefore, in the description using FIG. 7, the clustering unit 103 mainly describes processing for specifying the grid 1031 to which the content 1011 belongs.

  For the sake of simplicity, FIG. 7 shows an example in which the latitude and longitude of the position information of the content 1011 are each expressed as a 3-digit binary value. In this example, the N-ary value generation unit 101 expresses the latitude and longitude of the content 1011 as 3-digit binary values in the range of 000 to 111, and arranges these binary values one by one in order. A 6-digit binary value in the range of 000000 to 111111 is generated. Accordingly, the binary value “010101” is associated with the content 1011u having longitude 000 and latitude 111. Also, a binary value of “010110” is associated with the content 1011v whose longitude is 001 and latitude is 110. Further, a binary value of “010001” is associated with the content 1011w whose longitude is 000 and latitude is 101.

  On the other hand, in the illustrated example, the ground surface 1001 is divided into four upper grids 1041, and each upper grid 1041 is further divided into four grids 1031. That is, in the illustrated example, the grid has a quadtree structure. If the entire ground surface is the 0th layer grid of this quadtree structure, the upper grid 1041 is the 1st layer grid and the grid 1031 is the 2nd layer grid. As described above, in the clustering of this embodiment, the cluster 1021 and the grid 1031 correspond to each other. Therefore, in the illustrated example, the cluster also has a quadtree structure similar to the grid. Specifically, the cluster including the entire content 1011 on the ground surface 1001 corresponds to the root node of the quadtree structure, that is, the 0th layer, and the cluster including the content 1011 included in each upper grid 1041 is the first layer. And the cluster including the content 1011 included in each grid 1031 is the second layer cluster.

  Here, it will be described that “a plurality of contents 1011 having the same upper k digits of binary values generated by the N-ary value generation unit 101 are classified into the same cluster”.

  For example, content 1011 having longitudes 000 to 011 and latitudes 100 to 111 belongs to the upper grid 1041 in the upper left of the figure. In these ranges, the first digit of longitude is “0” and the first digit of latitude is “1”. Accordingly, the upper two digits of the binary value associated with the content 1011 are “01”, and the binary value of the content 1011 belonging to the upper grid 1041 is represented as “01xxxx”. Similarly, the binary values of the content 1011 belonging to other upper grids are expressed as “11xxxx”, “00xxxx”, and “10xxxx”. That is, regarding the upper grid 1041 that is the first layer of the quadtree structure grid, which upper grid 1041 the content 1011 belongs to is determined by the upper two digits of the binary value associated with the content 1011. . Therefore, the plurality of contents 1011 having the same upper two digits of the binary value belong to the same upper grid 1041 and are the same in the first cluster of the same cluster corresponding to the same upper grid 1041, that is, the quadtree structure cluster. Classified into clusters.

  Also, for example, content 1011 with longitude 000 or 001 and latitude 110 or 111 belongs to the grid 1031 at the upper left corner of the figure. In these ranges, the upper two digits of longitude are “00” and the upper two digits of latitude are “11”. Accordingly, the upper 4 digits of the binary value associated with the content 1011 are “0101”, and the binary value of the content 1011 belonging to the grid 1031 is represented as “0101xx”. Similarly, the binary value of the content 1011 belonging to another grid is also expressed in a form in which the upper 4 digits are specified. In other words, regarding the grid 1031 which is the second layer of the quadtree structure grid, the grid 1031 to which the content 1011 belongs is determined by the upper 4 digits of the binary value associated with the content 1011. Therefore, the plurality of contents 1011 having the same upper 4 digits of the binary value belong to the same grid 1031 and correspond to the same grid 1031, that is, the same cluster in the second layer of the quadtree structure cluster. are categorized.

  Furthermore, with reference to FIG. 7, the clustering of the contents 1011u to 1011w will be described as a specific example of clustering. First, the content 1011u is associated with the binary value “010101”. Since the upper 4 digits of this binary value are “0101”, it is determined that the content 1011u belongs to the grid 1031 at the upper left corner of the figure corresponding to the binary value represented by “0101xx”.

  Next, the content 1011v is associated with the binary value “010001”. Since the upper 4 digits of this binary value are “0100”, it is determined that the content 1011v belongs to the grid 1031 corresponding to the binary value represented by “0100xx”, that is, the grid 1031 different from the content 1011u. The

  Next, the content 1011w is associated with the binary value “010110”. Since the upper 4 digits of this binary value are “0101”, the content 1011v belongs to the grid 1031 at the upper left corner of the figure corresponding to the binary value represented by “0101xx”, that is, the same grid as the content 1011u. It is determined.

Here, a case is considered where the clustering sorting unit 105 sorts the contents 1011u to 1011w in the order of binary values. When the contents 1011u to 1011w are sorted in the order of binary values,
Content 1011v “010001”
Content 1011u “010101”
Content 1011w “010110”
It becomes. Content belonging to the same grid, that is, content 1011u and content 1011w that are classified into the same cluster are adjacent to each other in the sorting result. Therefore, the clustering unit 103 can specify the content 1011 classified into the same cluster 1021 from the result of sorting by the clustering sorting unit 105.

  Details of the processing of the clustering unit 103 and the clustering sort unit 105 will be described later.

(Merge part)
The merge unit 107 is realized by a CPU, a ROM, a RAM, and the like. The merge unit 107 includes a merge sort unit 109 and an adjacency determination unit 111 described later. The merge unit 107 sets the clusters 1021 that are determined to be adjacent to each other in a certain direction on the ground surface by the adjacent determination unit 111 as a processing target. The merge processing executed by the merge unit 107 includes search order merge and distance order merge as described later. In addition, the merge unit 107 may use the processing results of the merge sort unit 109 and the adjacency determination unit 111 in order to determine a cluster 1021 to be merged. Further, the merge unit 107 may refer to predetermined merge setting information stored in the storage unit 119 in order to set conditions for the merge process.

  The merge sort unit 109 is included in the merge unit 107 and is realized by a CPU, a ROM, a RAM, and the like. The merge sorting unit 109 sorts the clusters 1021 in a certain direction on the ground surface based on the result of the rank determination process based on latitude and longitude. The merge sorting unit 109 provides the sorting result to the adjacency determination unit 111. Here, in this embodiment, the rank determination process sorts the grid 1031 with respect to the direction on the ground surface, for example, the east-west direction, the north-south direction, the northwest-southeast direction, or the southwest-northeast direction, and the clusters included in each grid 1031. This may be a process of giving the sorting order of the grid 1031 as a rank to 1021.

  The adjacency determination unit 111 is included in the merge unit 107 and is realized by a CPU, a ROM, a RAM, and the like. The adjacency determination unit 111 determines whether or not clusters sorted in a certain direction on the ground surface by the merge sorting unit 109 are adjacent to each other in the direction. The adjacency determination unit 111 provides the determination result to the merge unit 107. Here, in the present embodiment, the adjacent determination process may be a process of determining whether or not the grid 1031 including the cluster 1021 is adjacent.

  Here, the function of the merge unit 107 will be described with reference to FIG. FIG. 8 is a diagram for explaining cluster merging according to the first embodiment of this invention. FIG. 8 shows grids 1031x and 1031y adjacent to each other in the longitude direction, and clusters 1021x and 1021y included in the respective grids.

  In the illustrated example, the merging unit 107 sets a cluster 1021 adjacent to the corresponding grid 1031 as a cluster 1021x and a cluster 1021y as a cluster 1021 adjacent to each other, and the clusters 1021 You may calculate the distance d between. The distance d between clusters may be, for example, the distance between the centers of the respective clusters 1021. The merge unit 107 merges the cluster 1021x and the cluster 1021y into a cluster 1021z when the distance d between the clusters is equal to or smaller than a predetermined threshold.

  As described above, the merge unit 107 may calculate the distance between the clusters 1021 when the clusters 1021 are adjacent to each other in the latitude or longitude direction. The merging unit 107 may calculate the distance between the clusters regardless of whether the clusters 1021 are adjacent to each other. Further, the merge unit 107 may store the clusters 1021 in the storage unit 119 as merge candidate clusters instead of immediately merging these clusters 1021 when the distance between the clusters is equal to or smaller than a predetermined threshold. In this case, the merging unit 107 then merges the clusters 1021 in order from the merge candidate cluster having the smallest distance between the clusters among the merge candidate clusters.

  Details of the processes of the merge unit 107, the merge sort unit 109, and the adjacency determination unit 111 will be described later.

(Input section)
Returning to FIG. 6, the input unit 113 is an example of an input device included in the information processing apparatus 100 according to the present embodiment. The input unit 113 is realized by, for example, a CPU, a ROM, a RAM, an input device, and the like. The input unit 113 converts a user operation performed on the keyboard, mouse, touch panel, and the like included in the information processing apparatus 100 into an electrical signal corresponding to the performed user operation, and generates an N-ary value generation unit 101 and a display control unit 115 is notified. For example, when an operation for instructing execution of clustering or an operation for instructing a change in the granularity of clustering is performed by the user, the input unit 113 generates information representing the instruction and generates an N-ary value generation unit 101. Output to etc.

(Display control unit)
The display control unit 115 is realized by a CPU, a ROM, a RAM, and the like, for example. For example, when the display control unit 115 receives a notification from the input unit 113 that a user operation for instructing display of the clustering result has been performed, the display control unit 115 is stored in the storage unit 119 or the like by the clustering unit 103 and the merge unit 107. The clustering result of the content 1011 is acquired. Thereafter, the display control unit 115 may perform display control such that, for example, an image of the clustering result as described with reference to FIG. 3 is configured and this image is displayed on the display unit 117 described later.

(Display section)
The display unit 117 is an example of a display device included in the information processing apparatus 100 according to the present embodiment. The display unit 117 is a display unit that displays various contents that can be executed by the information processing apparatus 100, execution screens of various applications, and the like. In addition, the display unit 117 may display various objects used for operating various content operations, execution states of various applications, and the like. Various information such as an image of the clustering result described with reference to FIG. 3 is displayed on the display screen of the display unit 117 under the control of the display control unit 115.

(Memory part)
The storage unit 119 is an example of a storage device included in the information processing apparatus 100 according to the present embodiment. The storage unit 119 may store various content data included in the information processing apparatus 100 and metadata associated with the content data. The storage unit 119 stores the binary value generated by the N-ary value generation unit 101, the result of the clustering unit 103 classifying the content 1011 into the cluster 1021, and the result of the merge unit 107 merging the cluster 1021. Also good. The storage unit 119 may also store execution data corresponding to various applications used by the display control unit 115 to display various types of information on the display unit 117. In addition, the storage unit 119 appropriately stores various parameters that need to be saved when the information processing apparatus 100 performs some processing, the progress of processing, or various databases. In the storage unit 119, each processing unit included in the information processing apparatus 100 according to the present embodiment can freely read and write.

(Supplementary information processing equipment)
Note that the information processing apparatus 100 according to the present embodiment may be an apparatus having a function of acquiring position information associated with the content 1011 from the content 1011 itself or an additional data file. Examples of the information processing device 100 include an imaging device such as a digital still camera and a digital video camera, a multimedia content viewer with a built-in storage device, a portable information terminal capable of recording, storing and browsing content, and a map service on a network. Content management browsing services, personal computer application software, portable game terminals having a photo data management function, mobile phones with cameras having a storage device, storage devices, digital home appliances and game machines having a photo data management function, and the like.

  Heretofore, an example of the function of the information processing apparatus 100 according to the present embodiment has been shown. Each component described above may be configured using a general-purpose member or circuit, or may be configured by hardware specialized for the function of each component. In addition, the CPU or the like may perform all functions of each component. Therefore, it is possible to appropriately change the configuration to be used according to the technical level at the time of carrying out the present embodiment.

  It should be noted that a computer program for realizing each function of the information processing apparatus according to the present embodiment as described above can be produced and installed in a personal computer or the like. In addition, a computer-readable recording medium storing such a computer program can be provided. The recording medium is, for example, a magnetic disk, an optical disk, a magneto-optical disk, a flash memory, or the like. Further, the above computer program may be distributed via a network, for example, without using a recording medium.

(1-3. Details of clustering and merge processing)
Next, with reference to FIGS. 9 to 29, details of the clustering process and the merge process in the first embodiment of the present invention will be described.

  FIG. 9 is a flowchart illustrating clustering and merging processing according to the first embodiment of this invention. As described with reference to FIG. 6, in the information processing apparatus 100 according to the present embodiment, the merging unit 107 performs the merging process on the cluster 1021 identified as a result of clustering by the clustering unit 103. Therefore, as a processing flow, clustering processing is first executed (step S101), and then merge related processing is executed (step S103). The merge related process is a generic name of processes in which related processes such as parameter setting are added to the merge process. Details of the clustering process and the merge-related process will be described later.

Table 1 is a table showing an example of combinations of grid levels set in the clustering process and merge thresholds set in the merge process. For example, when a set having a grid level of 10 and a merge threshold of 50 km is selected, a level 10 grid is used in the clustering process. Here, in the hierarchical structure of the grid described with reference to FIG. 2, the upper grid is divided into four for each level. Therefore, the level 10 grid is an overall 4 ¹⁰ divided grids of the earth surface in the range of level 0 grid. The merge threshold is a threshold for the distance d between clusters described with reference to FIG. In the above case, regarding a plurality of clusters 1021 included in the grid 1031 adjacent to each other, if the distance d between the clusters is 50 km or less, these clusters 1021 are merged.

  The grid level and the merge threshold as described above can be set to arbitrary values. The combination of the grid level and the merge threshold shown in Table 1 is an example of the combination of the grid level and the merge threshold such that the grid is included in a circle having a radius of the merge threshold near 40 degrees north latitude. . If the grid level is relatively increased with respect to the merge threshold (the grid size is reduced), it is determined that the grid is not adjacent at the time of determining the grid adjacency, and even if the distance between clusters included in the grid is less than or equal to the merge threshold The cluster may not be merged. However, if the range of grid adjacency determination can be expanded by using a 4-way search or higher level search described later, the amount of calculation increases as the grid size is reduced, but the result of clustering is more natural. Get closer to.

(Details of clustering process)
The clustering process by the clustering unit 103 will be described in more detail with reference to FIG. FIG. 10 is a diagram for explaining clustering in the first embodiment of the present invention. FIG. 10 shows a state where the sorting unit 105 for clustering sorts the content 1011 in the order of binary values generated by the N-ary value generation unit 101 and associated with the content 1011. As a result of the sorting by the clustering sorting unit 105, the contents 1011 are arranged in the order of clusters classified in the first layer cluster 1024 and the second layer cluster 1025. Note that FIG. 10 also shows an example in which the latitude and longitude of the position information of the content 1011 are each represented by a 3-digit binary value for the sake of simplicity.

As described above, in the clustering of the present embodiment, a plurality of contents 1011 having the same high-order k digits of binary values generated by the N-ary value generation unit 101 are classified into the same cluster. When k = 2 × m (m = 1, 2,...), The cluster into which the content 1011 having the same upper k digits of the binary value is classified is 2 ² = quadrant tree cluster. It becomes m-th layer. For example, a plurality of contents 1011 having the same upper two digits of the binary value are classified into the same cluster in the first layer of the quadtree structure. That is, the first-layer clusters of the quadtree structure are four clusters corresponding to binary values represented by “00xxxx”, “01xxxx”, “10xxxx”, or “11xxxx”, respectively. The second-layer cluster of the quadtree structure is 16 clusters corresponding to binary values represented by “0000xx”, “0001xx”, “0010xx”,... “1111xx”, respectively.

  In the example of FIG. 10, as an example of the first layer cluster 1024, a first layer cluster 1024a corresponding to “00xxx” and a first layer cluster 1024b corresponding to “01xxxx” are illustrated. As an example of the second layer cluster 1025, the second layer cluster 1025a corresponding to "0000xx", the second layer cluster 1025b corresponding to "0001xx", and the second layer cluster 1025c corresponding to "0010xx". , And a second layer cluster 1025d corresponding to “0011xx” is illustrated.

  For example, regarding the second layer cluster, the content 1011 associated with “000010” at the head of the sorted binary value is classified as the second layer cluster 1025a. Further, the content 1011 associated with the next four binary values “000100” to “000111” is classified into the second layer cluster 1025b. Further, the content 1011 associated with the next “001001” is classified into the second layer cluster 1025c, and the content 1011 associated with the next “001110” is classified into the second layer cluster 1025d.

  As for the first layer cluster, the content 1011 in which the seven binary values from “000010” to “001110” at the head of the sorted binary value are associated with each other is the first layer cluster. Classified as 1024a. Further, the contents 1011 associated with the following four binary values from “010011” to “011101” are classified into the first layer cluster 1024b.

  As described above, in the clustering according to the present embodiment, when the content 1011 is sorted in the order of the binary value generated by the N-ary value generation unit 101, the content 1011 is arranged in the cluster unit to be classified. That is, the clustering process in the present embodiment is realized by sorting the contents 1011 in the order of the binary values generated by the N-ary value generation unit 101.

Here, in general distance-based position clustering, a pair of content having a short distance is searched, so the number of processing times is the number of content combinations, and when the number of content is N, the number of processing times is O (N ² ). become. On the other hand, in the clustering in this embodiment, since the clustering process is substantially a sort process, if the number of contents is N, the number of processes is O (NlogN), and a smaller number of processes is sufficient. In addition, one process in general distance-based position clustering is a distance calculation between two-dimensional coordinates, whereas one process in clustering in this embodiment is a numerical comparison of numerical values. The load per process is lower.

  FIG. 11 is a diagram for describing cluster specifying information according to the first embodiment of this invention. FIG. 11 shows an arrangement of content specifying information for specifying the content 1011 and cluster specifying information for specifying the cluster 1021.

  In the present embodiment, the clustering unit 103 determines, based on the result of sorting by the clustering sorting unit 105, the first position where the content 1011 classified into a certain cluster 1021 appears and the number of content 1011 classified into this cluster 1021. Then, cluster specifying information for specifying the cluster 1021 is generated.

  In the illustrated example, the content 1011 is defined by a data structure called Item. Item can have the following data structure, for example.

struct Item {
uint32 id;
uint64 geocode;
};

  Here, id is a unique ID given to identify each content 1011. The geocode is a binary value generated by the N-ary value generation unit 101 for each content 1011. In the array of Items shown in the figure, each Item is sorted according to geocode.

  In the illustrated example, the cluster 1021 is defined by a data structure called Cluster. Cluster is, for example, the following data structure.

struct Cluster {
uint64_t clusterid;
uint32_t latcode;
uint32_t lngcode;
float latitude;
float longitude;
float halfEW;
float halfNS;
uint32 numLeaves;
Item * pLeaves;
};

  Here, clusterid is a unique ID given to identify each cluster 1021. Latcode and lngcode are latitude and longitude codes of the grid 1031 corresponding to the cluster 1021. For example, if the grid 1031 corresponding to the cluster 1021 corresponds to the binary value “100111”, the latcode is “011” and the lngcode is “101”. latitude, longitude, halfEW, and halfNS are information for defining the area of the cluster 1021.

  Further, numLeaves is the number of contents 1011 classified into the cluster 1021. * pLeaves is a pointer indicating the position of the first content 1011 classified into the cluster 1021 in the array in which the clustering sorting unit 105 sorts the items by geocode. These two elements are information for specifying the content 1011 classified into the cluster 1021. Similar to the example described with reference to FIG. 8, the Item array sorted by geocode is arranged in the order of clusters into which the content 1011 defined by Item is classified. Therefore, it is possible to specify the content 1011 classified into the cluster 1021 by the information “where (* pLeaves) to how many (numLeaves) in the array”.

  Here, in general clustering, when content is classified into clusters, information for identifying each classified content, such as a content ID array, is held as information defining the cluster. In this case, as the number of contents classified into clusters increases, the content ID array also increases, so the information defining the clusters also increases. On the other hand, in the clustering according to the present embodiment, the content 1011 classified into the cluster 1021 is specified by the information “from where in the array” as described above, and therefore the number of the content 1011 classified into the cluster 1021 increases. In addition, the size of information defining the cluster 1021 can be suppressed.

(Details of merge related processing)
FIG. 12 is a flowchart showing merge-related processing in the first embodiment of the present invention. In the merge related process, whether to perform the merge process is determined depending on whether the merge setting information (config) is set. When performing the merge process, the search is performed according to the contents of the merge setting information (config). A forward or distance order merge is performed. The merge setting information (config) is also used for setting parameters in search order merge and distance order merge.

  First, the merging unit 107 executes merge setting selection processing (step S201). In this merge setting selection process, merge setting information (config) is selected. Details of the merge setting selection process will be described later.

  Subsequently, the merge unit 107 determines whether data is set in the merge setting information (config) (step S203).

  If the merge setting information (config) is set in step S203, the merge unit 107 determines whether the distance order merge flag (sortPair) of the merge setting information (config) is “true”, that is, distance order merge is valid. It is determined whether or not (step S205).

  If the distance order merge flag (sortPair) in the merge setting information (config) is “false” in step S205, that is, if distance order merge is not valid, the merge unit 107 executes search order merge processing (step S205). S207). Details of the search order merge processing will be described later.

  On the other hand, if the distance order merge flag (sortPair) of the merge setting information (config) is “true” in step S205, that is, if the distance order merge is valid, the merge unit 107 executes distance order merge processing. (Step S209). Details of the distance order merge processing will be described later.

  On the other hand, if the merge setting information (config) is null in step S203, that is, no data is set, the merge unit 107 ends the process without executing the search order merge process and the distance order merge process. .

(Details of merge setting selection processing)
FIG. 13 is a diagram illustrating an example of merge setting information according to the first embodiment of this invention. FIG. 13 illustrates the maximum number of grids (maxGrid), search method (searchType), upper level search (upperLevel), distance order merge (sortPair), and distance calculation as elements of merge setting information.

  The merge setting information may be stored in the storage unit 119 of the information processing apparatus 100 in the form of a table as illustrated, for example. Each merge setting information is stored in the form of a merge setting record 1051, and the merge setting record 1051 may be identified by an index. Hereinafter, each element of the merge setting information will be described.

  The applied maximum number of grids (maxGrid) is the maximum number of grids for which merge setting information can be set. The merge setting information (config) used for the merge process is selected from merge setting information in which the number of grids is equal to or less than the maximum number of applied grids. Note that the number of grids referred to here may be the number of grids 1031 including the cluster 1021 as a result of the clustering process. Therefore, even when the grid level is large (the grid size is small), when the number of contents 1011 is small or the distribution is biased toward a specific grid, the merge setting information for the relatively small number of applied maximum grids. Can be selected.

  The search method (searchType) specifies a search method in the merge process. “Full Match” is a search method in which all combinations of the grid 1031 including the cluster 1021 are targeted for merge processing. The merge processing by this search method is hereinafter referred to as full match merge processing. “4 Dir”, “2 Dir”, and “1 Dir” represent a four-direction search, a two-way search, and a one-way search, respectively, and a search that uses a grid adjacent to a specific direction as a target of merge processing. It is a technique. These search methods are hereinafter referred to as neighborhood search merge processing. Details of the full match merge process and the neighbor search merge process will be described later.

  The upper level search (upperLevel) specifies whether or not to execute an upper level search and, if an upper level search is executed, up to which level the search is executed. “2” and “1” indicate that an upper level search is performed, and indicates an upper level search up to 2 levels higher and 1 level higher, respectively. “0” (disable) indicates that the upper level search is not performed. In the full match merging process, all combinations of the grid 1031 including the cluster 1021 are subjected to the merging process. Therefore, the upper level search is not necessary, and therefore the upper level search (upperLevel) is not defined. Details of the upper level search will be described later.

  The distance order merge flag (sortPair) specifies whether or not to execute distance order merge. The distance order merging is a merging method of merging sequentially from the pair of the grid 1031 including the cluster 1021 in order from the pair having the closest distance of the cluster 1021. When the distance order merge is executed, the clusters 1021 that are close to each other can be surely merged. However, since the grid pairs are temporarily stored, the storage capacity is occupied accordingly. “True” indicates that distance order merging is executed, and “false” indicates that search order merging described later is executed instead of distance order merging. Details of distance order merging will be described later.

  The distance calculation specifies a distance calculation method used when calculating the distance between clusters. When “great circle distance” is used, for example, assuming that the coordinates (longitude, latitude) of the centers of two clusters 1021 are (lon1, lat1) and (lon2, lat2), respectively, the distance d between the clusters is Calculated by Equation (1).

  When the “approximate great circle distance” is used, the distance d between clusters is calculated by the following formula (2) in the same case.

  Such merge setting information may be set in advance and stored in the storage unit 119, for example. The illustrated merge setting information is set such that the smaller the applied maximum number of grids (maxGrid) is, the higher the merge is, and the larger the applied maximum number of grids (maxGrid) is, the simpler the merge is. The merge setting information corresponding to the case where the number of grids exceeds 50000 is not defined. In this case, the merge process is not executed. This is because the load of merge processing increases as the number of grids increases, so the merge setting is changed according to the number of grids to adjust the maximum load of calculation.

  FIG. 14 is a flowchart showing merge setting selection processing according to the first embodiment of the present invention. In the merge setting selection process, merge setting information (config) used for the merge process is selected from the merge setting information described with reference to FIG. As described above, the merge setting information (config) may not be set, and it may be specified that the merge process is not executed.

  First, the merging unit 107 has a length of a grid list (glist) that is a list of grids 1031 including the cluster 1021, that is, the number of grids, the maximum number of grids applied (maxGrid) in the tail element of the merge setting list mlist. It is determined whether or not the following is true (step S301).

  In step S301, when it is determined that the number of grids is equal to or less than the maximum number of applied grids (maxGrid) in the tail element tail of the merge setting list mlist, the merge unit 107 sequentially performs the following on each element of the merge setting list mlist. Step S305 and step S307 are repeated (step S303: merge setting list loop).

  Step S305 is a step of comparing the length length of the grid list (glist) with the maximum number of applied grids (maxGrid) of the element of index i in the merge setting list mlist.

  In step S305, when it is determined that the length length of the grid list (glist) is equal to or less than the maximum number of applied grids (maxGrid), the merge unit 107 includes the index i of the merge setting list mlist in the merge setting information (config). Elements are set (step S307). When merge setting information (config) that has already been set exists, it is overwritten with the element of index i of the newly set merge setting list mlist.

  On the other hand, if it is determined in step S305 that the length length of the grid list (glist) exceeds the maximum number of applied grids (maxGrid), the setting of the merge setting information (config) is not changed, and the merge setting list loop is executed. To continue.

  When the merge setting list loop in step S303 ends, the merge unit 107 ends the merge setting selection process.

  On the other hand, when it is determined in step S301 that the number of grids exceeds the maximum number of applicable grids (maxGrid) in the tail element of the merge setting list mlist, the merge unit 107 sets null in the merge setting information (config). (Step S309). null indicates a state in which no data is set. As described above, when the merge setting information (config) is null, the merge unit 107 does not execute subsequent merge processing.

(Details of search order merge processing)
The search order merging process includes a process of searching for clusters 1021 whose distance between each other is equal to or smaller than a predetermined threshold, and sequentially merging such clusters 1021 in the searched order. In the full match merge process described later, the distance between the clusters 1021 is calculated for all combinations of the grid 1031 including the cluster 1021. Further, in the neighborhood search merge process described later, the grid 1031 including the cluster 1021 is sorted in a specific direction, and the distance between the clusters 1021 included in the grid 1031 adjacent to the direction is calculated.

  FIG. 15 is a flowchart showing search order merge processing according to the first embodiment of the present invention. In the search order merge, either the full match merge process or the neighbor search merge process is executed according to the contents of the merge setting information (config). Further, when the neighborhood search merge process is executed, it is executed with or without an upper search according to the contents of the merge setting information (config).

  First, the merge unit 107 determines whether or not the search method (searchType) of the merge setting information (config) is “Full Match” (step S401).

  If the search method (searchType) of the merge setting information (config) is “Full Match” in step S401, the merge unit 107 executes a full match merge process (step S403). Further details of the full match merge process will be described later.

  On the other hand, when the search method (searchType) of the merge setting information (config) is not “Full Match” in step S401, the merge unit 107 further performs an upper level search (upperLevel) of the merge setting information (config), It is determined whether or not it is “0” (step S405).

  If the upper level search (upperLevel) of the merge setting information (config) is “0” in step S405, the merge unit 107 executes a neighbor search merge process without an upper search (step S407). The details of the neighborhood search merging process without upper search will be described later.

  On the other hand, if the upper level search (upperLevel) of the merge setting information (config) is not “0” in step S405, the merge unit 107 executes a neighbor search merge process with an upper search (step S409). The details of the neighborhood search merging process with upper search will be described later.

  As described above, the merge unit 107 executes any one of the full match merge process, the neighborhood search merge process without upper search, or the neighbor search merge process with upper search, and ends the search order merge process.

(Details of full match merge processing)
FIG. 16 is a flowchart showing the full match merge process according to the first embodiment of the present invention. In the full match merge process, all combinations of the grid 1031 including the cluster 1021 are targeted for search.

  First, in step S501, the merging unit 107 sequentially repeats the following step S503 for each element of the grid list (glist) that is a list of the grid 1031 including the cluster 1021 (grid loop for merging). Note that the element of the grid list (glist) that is the processing target in the loop of step S501 is the element of the index i of the grid list (glist).

  In step S503, the merge unit repeats the following steps S505 to S509 in order starting from the element that is the element next to the element of the index i that is currently processed in the grid list (glist) (a grid loop to be merged). Note that the element of the grid list (glist) that is the processing target in the loop of step S503 is the element of the index j of the grid list (glist).

  In subsequent step S505, the merging unit 107 calculates the distance d between the element at index i (grid to be merged) and the element at index j (grid to be merged) in the grid list (glist). To do. Here, the distance between the element of index i and the element of index j is, for example, the center between cluster 1021 included in grid 1031 that is an element of index i and cluster 1021 included in grid 1031 that is an element of index j. It may be a distance.

  The subsequent step S507 is a step of comparing the distance d calculated in step S505 with a predetermined threshold th. Here, the threshold th may be, for example, the merge threshold described with reference to Table 1.

  If it is determined in step S507 that the distance d is equal to or smaller than the threshold th, the merge unit 107 merges the element at index i and the element at index j in the grid list (glist) (merge: step S509). ). Here, the merging unit 107 merges the cluster 1021 corresponding to the grid 1031 that is the element of the index i in the grid list (glist) and the cluster 1021 corresponding to the grid 1031 that is the element of the index j. And the new cluster 1021 is associated with the grid 1031 associated with each cluster 1021 before merging. That is, after this merging, the cluster 1021 corresponding to the grid 1031 that is the element of the index i and the cluster 1021 corresponding to the grid 1031 that is the element of the index j both become this new cluster 1021.

  On the other hand, when it is determined in step S507 that the distance d exceeds the threshold th, the merging unit 107 increments the index j without merging the elements of the index i and the index j of the grid list (glist). Then, the next element of the grid list (glist) is referred to (step S503).

  When the element of the grid list (glist) is referred to the end by the loop of step S503, the merge unit 107 increments the index i and refers to the next element of the grid list (glist) (step S501).

  When the elements of the grid list (glist) are referred to the end by the loop of step S501, the merge unit 107 ends the full match merge process.

  The processing in step S509 is an example of specific processing when the merge unit 107 is implemented as software, for example. In this case, the merge unit 107 refers to the single grid list (glist) stored in the storage unit 119 using both the index i and the index j, and sequentially updates the contents of the grid list (glist). If the merge unit 107 is implemented by software other than the above, and if the merge unit 107 is implemented by software having a different specification from the above example, if the substantial processing content is in accordance with the above flowchart, For example, the method of referring to the grid 1031 to be processed and the timing for reflecting the merge of clusters in the data can be designed as appropriate.

(Details of neighbor search merge processing)
Next, the neighborhood search merge process in this embodiment will be described. In the neighborhood search merging process, the merging sort unit 109 sorts the clusters 1021 in a certain direction on the ground surface based on the result of the first rank determination process based on the latitude and longitude. Further, the adjacency determination unit 111 determines whether or not the clusters sorted in the direction are adjacent to each other in the direction. Further, the merging unit 107 calculates the distance between the clusters for the clusters determined to be adjacent to each other in this direction. The merging unit 107, the merging sorting unit 109, and the adjacency determining unit 111 may merge clusters in the same process for other directions on the ground surface.

  Here, in the present embodiment, the direction on the ground surface can be, for example, the horizontal direction, the vertical direction, the diagonally lower right direction, or the diagonally upper right direction. The merge unit 107 including the merge sort unit 109 may perform the following neighborhood search merge process for one of these directions. The neighborhood search in this case is also called a one-way search. Further, the merging unit 107 including the merging sort unit 109 may execute the following neighborhood search merging process for two of the above directions. The neighborhood search in this case is also called a two-way search. Further, the merge unit 107 including the merge sort unit 109 may execute the following neighborhood search merge process in the above four directions. The neighborhood search in this case is also called a four-way search.

  Further, in the present embodiment, the merge sorting unit 109 sorts the sorting order given by sorting the grid 1031 in a certain direction on the ground surface into the clusters 1021 included in each grid 1031 as the first order determination processing. It may be given as a ranking. As described with reference to FIG. 1, in the present embodiment, the grid 1031 and the cluster 1021 included in the grid 1031 correspond one-to-one. Therefore, the sort order given to the grid can be given as it is as the rank of the cluster. Since the grid 1031 is an area defined by a known boundary, it is easy to sort in a specific direction on the ground surface. Therefore, it is possible to speed up the sorting of the cluster 1021 by the configuration of the present embodiment that gives the rank to the cluster 1021 based on the sorting result of the grid 1031.

  FIG. 17A to FIG. 17D are diagrams illustrating directions in which a grid search is executed in the neighborhood search merge process according to the first embodiment of the present invention. In the neighbor search merge process, a grid list (glist) that is a list of grids 1031 including the cluster 1021 is sorted in a specific direction, and the grids 1031 adjacent to each other in the sorted grid list (glist) are merged. The target of. 17A to 17D show indexes given to the grids when the sorting order of the grid list (glist) is defined based on the respective directions. 17A to 17D, the arrangement of the grid itself is the same.

  FIG. 17A is a diagram illustrating a case where a search for a grid list (glist) is performed in the horizontal direction. The horizontal direction on the ground surface 1001 is also called the east-west direction. In the following drawings, the longitude direction is illustrated as the x-axis direction, and the latitude direction is illustrated as the y-axis direction. Here, the horizontal direction is the x-axis direction.

  In this case, the grid list (glist) is sorted in the horizontal direction. The sort order of the grid list (glist) is determined, for example, by defining the following relationship between two grids whose coordinates are (x1, y1) and (x2, y2) as follows.

if (y2 ≠ y1)
Judgment based on the magnitude of y1 and y2 (smaller is before)
else
Judgment based on the magnitude of x1 and x2 (the smaller one is before)

  FIG. 17B is a diagram illustrating a case where a search for a grid list (glist) is performed in the vertical direction. The vertical direction on the ground surface 1001 is also called the north-south direction. In this case, the grid list (glist) is sorted in the vertical direction. The sort order of the grid list (glist) is determined, for example, by defining the following relationship between two grids whose coordinates are (x1, y1) and (x2, y2) as follows.

if (x2 ≠ x1)
Judgment based on the magnitude of x1 and x2 (the smaller one is before)
else
Judgment based on the magnitude of y1 and y2 (smaller is before)

  FIG. 17C is a diagram illustrating a case where a search for a grid list (glist) is executed in a diagonally lower right direction. The diagonally lower right direction on the ground surface 1001 is also called the northwest-southeast direction. In this case, the grid list (glist) is sorted in the diagonally lower right direction. The sort order of the grid list (glist) is determined, for example, by defining the following relationship between two grids whose coordinates are (x1, y1) and (x2, y2) as follows.

sum1 = x1 + y1
sum2 = x2 + y2
if (sum1 ≠ sum2)
Judgment based on size of sum1, sum2 (smaller is before)
else
Judgment based on y1 and y2 (larger is before)

  FIG. 17D is a diagram illustrating a case where a search for a grid list (glist) is executed in a diagonally upper right direction. The diagonally upper right direction on the ground surface 1001 is also called the southwest-northeast direction. In this case, the grid list (glist) is sorted in the diagonally upper right direction. The sort order of the grid list (glist) is determined, for example, by defining the following relationship between two grids whose coordinates are (x1, y1) and (x2, y2) as follows.

The distance from the maximum y coordinate to y1, y2 is defined as y1 ′, y2 ′. Sum1 = x1 + y1 ′
sum2 = x2 + y2 ′
if (sum1 ≠ sum2)
Judgment based on size of sum1, sum2 (smaller is before)
else
Judgment based on the magnitude of y1 and y2 (smaller is before)

  18A to 18C are diagrams for describing a one-way search, a two-way search, and a four-way search in the neighborhood search merge process according to the first embodiment of the present invention. In the neighborhood search merging process according to the present embodiment, three types of searches, that is, a one-way search, a two-way search, and a four-way search, in which the grid search directions described with reference to FIGS. 17A to 17D are selected or combined. The method is set.

  FIG. 18A is a diagram illustrating a one-way search. In the one-way search, only one grid search direction is selected. In the illustrated example, the horizontal direction is selected. In this case, two grids that are adjacent to a certain grid in the horizontal direction are to be merged. Since the search direction is one, the grid list (glist) may be sorted once in the merge process. Note that the grid search direction to be selected is not limited to the horizontal direction, and may be any one of the vertical direction, the diagonally lower right direction, and the diagonally upper right direction. In such a one-way search, the number of sorts is one, and the maximum number of distance calculations in the merge process is approximately N (times), where N is the number of clusters.

  FIG. 18B is a diagram illustrating a two-way search. In the two-way search, two grid search directions are combined. In the illustrated example, the horizontal direction and the vertical direction are combined. In this case, four grids adjacent to a certain grid in the horizontal direction or the vertical direction are subjected to the merge process. Since there are two search directions, the grid list (glist) is sorted twice in the merge process. Note that the grid search directions to be combined are not limited to the horizontal direction and the vertical direction, and instead of these directions, either the diagonally lower right direction or the diagonally upper right direction, or both may be combined. In such a two-way search, the number of times of sorting is 2, and the maximum number of distance calculations in the merge process is about 2N (times), where N is the number of clusters.

  FIG. 18C is a diagram illustrating a four-way search. In the four-direction search, four grid search directions are combined. In the present embodiment, the horizontal direction, the vertical direction, the diagonally lower right direction, and the diagonally upper right direction are combined. In this case, a grid located in the vicinity in a horizontal direction, a vertical direction, or an oblique direction with respect to a certain grid is a target of the merge process. Here, considering that the distance to the adjacent grid in the diagonal direction is longer than the grid adjacent in the horizontal direction or the vertical direction, the horizontal direction and the vertical direction are adjacent to each other as shown in the figure. The grids up to the grid may be subject to merge processing as “adjacent grids”. In this case, 12 grids in the vicinity of the horizontal direction, the vertical direction, and the diagonal direction are targeted for the merge process. In such a four-way search, the number of sorts is four, and the maximum number of distance calculations in the merge process is about 4N (times), where N is the number of clusters.

  Note that the maximum number of distance calculations for the one-way search, the two-way search, and the four-way search is O (N). In addition, since the sorting is fast and O (NlogN), it can be considered that if the number of clusters becomes enormous, the sorting process becomes dominant in terms of computational complexity.

  FIG. 19 is a flowchart showing neighborhood search merge processing (no upper search) in the first embodiment of the present invention. The neighborhood search merge process (with upper search) will be described later together with details of the upper search.

  First, the merging unit 107 sets the direction (dir) to “horizontal” and executes an adjacent search process (no upper search) (step S601). The neighbor search process (without upper search) will be described later.

  Subsequently, the merge unit 107 determines whether the search method (searchType) of the merge setting information (config) is “2 Dir” or “4 Dir” (step S603). Here, when the search method (searchType) of the merge setting information (config) is neither “2 Dir” nor “4 Dir”, the merge unit 107 determines that a one-way search is designated and ends the process. To do.

  If the search method (searchType) is “2 Dir” or “4 Dir” in step S603, the merge unit 107 sets the direction (dir) to “vertical” and performs an adjacent search process (no upper search). Is executed (step S605).

  Subsequently, the merge unit 107 determines whether the search method (searchType) of the merge setting information (config) is “4 Dir” (step S607). Here, when the search method (searchType) of the merge setting information (config) is not “4 Dir”, the merge unit 107 determines that a two-way search is designated, and ends the process.

  If the search method (searchType) is “4 Dir” in step S607, the merge unit 107 determines that a four-way search is designated. In this case, the merging unit 107 executes the adjacent search process (no upper search) with the direction (dir) as “diagonal lower right” (step S609), and then performs the adjacent search process (upper right with the direction as “oblique upper right”). (No search) is executed (step S611), and the process is terminated.

  FIG. 20 is a flowchart showing the adjacent search process (no upper search) in the first embodiment of the present invention. The adjacent search process (with upper search) will be described later together with details of the upper search.

  First, the adjacency determination unit 111 determines whether the search method (searchType) of the merge setting information (config) is “4 Dir” and the direction (dir) is “horizontal” or “vertical”. (Step S701).

  Here, in step S701, when the search method (searchType) of the merge setting information (config) is “4 Dir” and the direction (dir) is “horizontal” or “vertical”, the adjacency determining unit 111 Then, 2 is set as the adjacency determination threshold th_n (step S703). In the present embodiment, the adjacency determination threshold th_n can be set in units of grid intervals. The adjacency determination threshold th_n is a threshold used for adjacency determination between grids to be described later, and can be set separately from a predetermined threshold th used for merge determination.

  On the other hand, when the search method (searchType) of the merge setting information (config) is not “4 Dir” in step S701, that is, “1 Dir” or “2 Dir”, or the direction (dir) is “horizontal”. However, if it is not “vertical”, the adjacency determination unit 111 sets 1 as the adjacency determination threshold th_n (step S705). In the present embodiment, the adjacent search process is executed when the search method (searchType) of the merge setting information (config) is “1 Dir”, “2 Dir”, or “4 Dir”.

  Here, the adjacency determination threshold th_n is set according to the search method (searchType) of the merge setting information (config) as described with reference to FIGS. 18A to 18C. Then, a grid adjacent to a grid in the horizontal direction or the vertical direction (distance between grids is one grid) is subject to merge processing, whereas in a four-way search, the grid is horizontal This corresponds to the fact that up to two adjacent grids in the vertical direction (the distance between the grids is equivalent to two grids) can be the target of merge processing as “adjacent grids”.

  Subsequently, the merge sort unit 109 temporarily sorts the grid list (glist) in the direction (dir) (tmpSort) (step S707). Here, the direction (dir) is specified as a parameter when the adjacent search process (no upper search) is executed in, for example, step S601, step S605, step S609, and step S611 in FIG. As described with reference to FIGS. 17A to 17D, the temporary sorting (tmpSort) can be a process of temporarily giving an index to each grid of the grid list (glist). Specifically, for example, when the direction (dir) is “horizontal”, an index as described with reference to FIG. 17A is temporarily given to each grid of the grid list (glist). When the direction (dir) is “vertical”, an index as described with reference to FIG. 17B is temporarily given to each grid of the grid list (glist). When the direction (dir) is “diagonally lower right”, an index as described with reference to FIG. 17C is temporarily given to the grid of the grid list (glist). When the direction (dir) is “diagonally upper right”, the grid as described with reference to FIG. 17D is temporarily given to the grid of the grid list (glist). A loop process in step S709 described later is executed according to the temporarily given index. After that, the index of the grid list (glist) is returned to the stored original index.

  In subsequent step S709, the merging unit 107 and the adjacency determining unit 111 repeat the following steps S711 to S717 in order from the top element of the grid list (glist) according to the index set in step S707 (grid loop to be merged). . Note that the element of the grid list (glist) that is the processing target in the loop of step S709 is the element of the index i of the grid list (glist).

  In step S711, the adjacency determination unit 111 determines whether the element of index i in the grid list (glist) is adjacent to the next element (element of index i + 1). Here, for example, when the direction (dir) is “horizontal”, the adjacent positions between the grids in the present embodiment are the same in the vertical position of each grid 1031 (y coordinate in the example of FIG. 17A). Further, it can be determined by whether or not the difference in the horizontal position (x coordinate in the example of FIG. 17A) is equal to or smaller than the adjacency determination threshold th_n. Therefore, the determination process of adjacent grids in step S711 is a process with a small processing load compared to a process including an operation such as calculation of a distance between arbitrary coordinates.

  If it is determined in step S711 that the element of index i in the grid list (glist) and the next element (element of index i + 1) are adjacent to each other, the merge unit 107 determines that the grid list (glist ) The distance d calculation process (distance) between the element at index i and the element at index i + 1 is executed (step S713). Here, the distance d between the element of index i and the element of index i + 1 is, for example, the cluster 1021 included in the grid 1031 that is the element of index i and the cluster 1021 included in the grid 1031 that is the element of index i + 1. And the center distance.

  The subsequent step S715 is a step of comparing the distance d calculated in step S713 with a predetermined threshold th. Here, the predetermined threshold th may be, for example, the merge threshold described with reference to Table 1.

  If it is determined in step S715 that the distance d is equal to or smaller than the threshold th, the merge unit 107 merges the element at index i and the element at index i + 1 in the grid list (glist) (merge: step S717). ). Here, the merging unit 107 merges the cluster 1021 corresponding to the grid 1031 that is an element of the index i of the grid list (glist) and the cluster 1021 corresponding to the grid 1031 that is the element of the index i + 1. And the new cluster 1021 is associated with the grid 1031 associated with each cluster 1021 before merging. That is, after this merging, the cluster 1021 corresponding to the grid 1031 that is the element of the index i and the cluster 1021 corresponding to the grid 1031 that is the element of the index i + 1 both become the new cluster 1021.

  On the other hand, when it is determined in step S711 that the element of index i in the grid list (glist) and the next element (element of index i + 1) are not adjacent to each other, the merging unit 107 performs steps S713 to S717. Is skipped, the index i is incremented, and the next element of the grid list (glist) is referred to (step S709). In this way, by determining whether or not to calculate the distance d with a relatively large processing load (step S713) by the adjacent determination processing between grids with a relatively small processing load (step S711), the processing is performed. It becomes possible to reduce the overall load.

  Further, in the above adjacent search processing, since the distance between clusters is calculated only between adjacent grids in the sorted grid list (glist), the maximum number of calculations for a certain direction (dir) is the number of clusters. If N is N, N-1 (times) is obtained. However, the number of times of sort processing is added to this as O (NlogN) as described above.

(Details of upper search processing)
FIG. 21 is a flowchart showing neighborhood search merge processing (with upper search) in the first embodiment of the present invention.

  First, the merge unit 107 generates an upper grid list (ulist) (step S801). The upper grid list (ulist) generated here will be described with reference to FIG. 22 and FIG.

  22 and 23 are diagrams for explaining the upper grid list generated in the upper search in the first embodiment of the present invention. Here, a case where the upper search is executed in the horizontal direction will be described as an example. It should be noted that the upper search can be executed in the same manner in the vertical direction, the diagonally lower right direction, and the diagonally lower right direction.

  As shown in FIG. 22, in this example, the grids numbered 0 to 11 belong to the six upper grids I to VI, respectively. As shown in FIG. 23, the grid and the upper grid having such an arrangement correspond to the heads of the grids belonging to the upper grids I to VI from the grid list (glist) in which the grids 0 to 11 are arranged. A list obtained by extracting grids (the grids belonging to each upper grid are sorted in the horizontal direction) is an upper grid list (ulist). As will be described later, the upper grid list (ulist) can be sorted in the horizontal direction, for example.

  Returning to FIG. 21, the merging unit 107 next performs a merging process in the upper grid (step S <b> 803). The merge processing in the upper grid executed here will be described with reference to FIG.

  FIG. 24 shows an example of merge processing in the upper grid in the upper grid of I in the example of FIG. As shown in FIG. 24, one set of round robin combinations of each grid belonging to the upper grid of I is targeted. The maximum number of times of distance calculation in the merge processing in the upper grid is about N / 4 × 6 = 1.5 N (times), where N is the number of clusters.

  Returning to FIG. 21, the merging unit 107 then performs the adjacent search process (with upper search) with the direction (dir) as “horizontal” (step S <b> 805). The neighbor search process (with upper search) will be described later.

  Subsequently, the merge unit 107 determines whether the search method (searchType) of the merge setting information (config) is “2 Dir” or “4 Dir” (step S807). Here, when the search method (searchType) of the merge setting information (config) is neither “2 Dir” nor “4 Dir”, the merge unit 107 determines that a one-way search is designated and ends the process. To do.

  If the search method (searchType) is “2 Dir” or “4 Dir” in step S807, the merge unit 107 sets the direction (dir) to “vertical” and performs an adjacent search process (with upper search). Is executed (step S809).

  Subsequently, the merge unit 107 determines whether or not the search method (searchType) of the merge setting information (config) is “4 Dir” (step S811). Here, when the search method (searchType) of the merge setting information (config) is not “4 Dir”, the merge unit 107 determines that a two-way search is designated, and ends the process.

  If the search method (searchType) is “4 Dir” in step S811, the merge unit 107 determines that a four-way search is designated. In this case, the merging unit 107 executes the adjacent search process (with upper search) with the direction (dir) as “diagonally lower right” (step S813), and then sets the direction as “oblique upper right” with the adjacent search process (upper search). Search is performed) (step S815), and the process is terminated.

  FIG. 25 is a flowchart showing an adjacent search process (with upper search) in the first embodiment of the present invention.

  First, the adjacency determination unit 111 determines whether the search method (searchType) of the merge setting information (config) is “4 Dir” and the direction (dir) is “horizontal” or “vertical”. (Step S901).

  Here, in step S901, when the search method (searchType) of the merge setting information (config) is “4 Dir” and the direction (dir) is “horizontal” or “vertical”, the adjacency determining unit 111 Then, 2 is set as the adjacency determination threshold th_n (step S903). In the present embodiment, the adjacency determination threshold th_n may be set with the upper grid interval as a unit. The upper grid interval is twice the grid interval. The adjacency determination threshold th_n is a threshold used for adjacent determination between upper grids, which will be described later, and can be set separately from a predetermined threshold th used for merge determination.

  On the other hand, if the search method (searchType) of the merge setting information (config) is not “4 Dir” in step S901, that is, “1 Dir” or “2 Dir”, or the direction (dir) is “horizontal”. However, if it is not “vertical”, the adjacency determination unit 111 sets 1 as the adjacency determination threshold th_n (step S905). In the present embodiment, the adjacent search process is executed when the search method (searchType) of the merge setting information (config) is “1 Dir”, “2 Dir”, or “4 Dir”.

  Here, the adjacency determination threshold th_n is set according to the search method (searchType) of the merge setting information (config) as described with reference to FIGS. 18A to 18C. Then, a grid adjacent to a grid in the horizontal direction or the vertical direction (distance between grids is one grid) is subject to merge processing, whereas in a four-way search, the grid is horizontal This corresponds to the fact that up to two adjacent grids in the vertical direction (the distance between the grids is equivalent to two grids) can be the target of merge processing as “adjacent grids”.

  Subsequently, the merge sorting unit 109 sorts the upper grid list (ulist) with respect to the direction (dir) as described with reference to FIG. 23 (sort) (step S907). Here, the direction (dir) is specified as a parameter when the adjacent search process (with upper search) is executed in, for example, step S805, step S809, step S813, and step S815 in FIG. The sort (sort) is a process of giving an index to each grid of the grid list (glist) as described with reference to FIGS. 17A to 17D, and the top grid of each upper grid included in the upper grid list (ulist). Can be executed against.

  In the subsequent step S909, the merge unit 107 and the adjacency determination unit 111 repeat the following steps S913 to S923 in order from the top element of the upper grid list (ulist) according to the sorting result in step S907 (upper grid list loop). . Note that the element of the upper grid list (ulist) that is the processing target in the loop of step S909 is the element of the index i of the upper grid list (ulist).

  In step S911, the adjacency determination unit 111 determines whether the upper grid corresponding to the element of index i in the upper grid list (ulist) and the upper grid corresponding to the next element (element of index i + 1) are adjacent to each other. Determine whether. Here, for example, when the direction (dir) is “horizontal”, the adjacent positions between the upper grids in this embodiment are the same in the vertical position of each grid 1031 (y coordinate in the example of FIG. 17A). In addition, it can be determined by whether or not the difference in the horizontal position (x coordinate in the example of FIG. 17A) is equal to or smaller than the adjacency determination threshold th_n. Therefore, the process for determining the adjacency between the upper grids in step S911 is a process with a smaller processing load compared to a process including an operation such as calculation of a distance between arbitrary coordinates.

  Here, in step S911, it is determined that the upper grid corresponding to the element of index i in the upper grid list (ulist) and the upper grid corresponding to the next element (element of index i + 1) are adjacent to each other. In step S913, the merging unit 107 performs the following steps for each of the grids (subgrids) belonging to the upper grid (first upper grid) corresponding to the element of the index i of the upper grid list (ulist). S915 to S921 are repeated (sub grid loop of the first upper grid). Here, it is assumed that a grid (subgrid) to be processed in the loop of step S913 is a.

  Further, in step S915, the merging unit 107 performs the following steps for each of the grids (subgrids) belonging to the upper grid (second upper grid) corresponding to the element of the index i + 1 of the upper grid list (ulist). S917 to S921 are repeated (second grid grid sub-grid loop). Here, b is a grid (subgrid) that is a processing target of the loop in step S915.

  In step S917, the merge unit 107 executes a calculation process (distance) of the distance d between the grid a and the grid b (step S917). Here, the distance d between the grid a and the grid b may be the center distance between the cluster included in the grid a and the cluster included in the grid b, for example.

  The subsequent step S919 is a step of comparing the distance d calculated in step S917 with a predetermined threshold th. Here, the predetermined threshold th may be, for example, the merge threshold described with reference to Table 1.

  Here, when it is determined in step S919 that the distance d is equal to or smaller than the threshold th, the merge unit 107 merges the cluster included in the grid a and the cluster included in the grid b (merge: step S921). . Here, the merge unit 107 generates a new cluster by merging the cluster corresponding to the grid a and the cluster corresponding to the grid b, and associates the new cluster with both the grid a and the grid b. That is, after this merging, both the cluster corresponding to the grid a and the cluster corresponding to the grid b become this new cluster.

  On the other hand, when it is determined in step S911 that the upper grid corresponding to the element of index i in the upper grid list (ulist) and the upper grid corresponding to the next element (element of index i + 1) are not adjacent to each other. The merging unit 107 skips the processes in steps S913 to S921, increments the index i, and refers to the next element in the upper grid list (ulist) (step S909). In this way, by determining whether or not to calculate the distance d with a relatively large processing load (step S917) by the adjacent determination processing between the upper grids with a relatively small processing load (step S911), It becomes possible to reduce the load of the entire processing.

  Here, with reference to FIG. 26, the process of said step S913-S921 is further demonstrated. FIG. 26 shows steps S913 to S921 when the direction (dir) is “horizontal” and the upper grids of I and III in the example of FIG. 22 become the first upper grid and the second upper grid, respectively. An example of the process is shown.

In the illustrated example, the grids that are the targets of the loop processing (grid a) in step S913 are the grids belonging to the upper I grid. In addition, the grids that are the targets of the loop processing in step S915 are the grids belonging to the upper grid of III. Therefore, the combinations of the grids that are the targets of the merge processing between the grid a and the grid b in steps S917 to S921 are the grids belonging to the upper grid of I and the higher rank of III as shown in FIG. The total number of combinations with each grid belonging to the grid is 6 sets. The maximum number of distance calculations in the merge process between such upper grids is approximately N / 4 × 4 ² = 4N (times), where N is the number of clusters.

  FIG. 27 shows a target grid (light thin) of merge processing for a certain grid (grid represented by dark hatching) when neighborhood search merge processing (with high-order search) is executed in the first embodiment of the present invention. It is a figure which shows the grid) represented by hatching. FIG. 27 shows an example in which the search type (searchType) of the merge setting information (config) is “4 Dir” and the upper level search (upperLevel) is 1.

  In this case, twelve upper grids (represented by bold lines) adjacent to each other in the four directions are merge processing targets in the upper search. Further, the three grids in the upper grid to which the certain grid belongs belong to the merge process in the upper grid. Therefore, the grids that are the targets of the merge process by the neighbor search merge process including the upper search process are 51 grids in total. Since it is a four-way search, the number of sorting is four, and the maximum distance calculation number is (4 × 4N) + 1.5N = 17.5N (times), where N is the number of clusters.

  The upper search process as described above may be executed by a one-way search or a two-way search, but is preferably executed by a four-way search in terms of the shape of the search range in the present embodiment. Further, the number of sorts does not change depending on whether the upper search process is executed or not, but the number of distance calculations increases.

(Details of distance order merge processing)
The distance order merging process searches for a cluster 1021 whose distance between each other is a predetermined threshold or less, stores such a cluster 1021 as a merge candidate cluster, and among the stored merge candidate clusters, Merging in order from the merge candidate cluster with the smallest distance.

  FIG. 28 is a diagram for describing an overview of distance order sorting in the first embodiment of the present invention. In FIG. 28, clusters 1021 s to 1021 u included in the three grids 1031 adjacent to each other in the horizontal direction (the search direction in this example) are illustrated. Here, it is assumed that the center distance between the cluster 1021s and the cluster 1021t is d1, and the center distance between the cluster 1021t and the cluster 1021u is d2. Both d1 and d2 are below a predetermined threshold th for the merge process, and d1> d2.

  Here, when the adjacent search process as described above is executed in the direction shown in the drawing, the target of the merge process is the combination of the cluster 1021s and the cluster 1021t. Since the center distance d1 between the cluster 1021s and the cluster 1021t is less than the predetermined threshold th for the merge process, the cluster 1021s and the cluster 1021t are merged to become the cluster 1021v. Here, it is assumed that the center distance between the cluster 1021v and the cluster 1021u is d3. It is assumed that d3 is larger than d1 and d2 and exceeds a predetermined threshold th for merge processing.

  Thereafter, the cluster 1021v and the cluster 1021u are to be merged. However, since the center distance d3 between the cluster 1021v and the cluster 1021u exceeds a predetermined threshold th for the merge process, the cluster 1021v and the cluster 1021u are not merged, and the clusters 1021s to 1021u are merged. The cluster 1021w is not generated.

  Thus, when merge processing is performed in the search order, the center distance between clusters is smaller in d2 between the cluster 1021t and the cluster 1021u than in d1 between the cluster 1021s and the cluster 1021t. Regardless, there may be a problem that the cluster 1021s and the cluster 1021t are merged and the cluster 1021u is not merged.

  In order to solve such a problem, distance order merging is performed. In this embodiment, as described above, the merge order information (config) includes the distance order merge flag (sortPair), and when the distance order merge flag (sortPair) is “true”, the distance order merge is executed. The

  FIG. 29 is a flowchart showing distance order merge processing according to the first embodiment of the present invention.

  First, the merge unit 107 performs the search order merge process described with reference to FIG. 15 (step S1001). However, in full match merge processing or neighborhood search merge processing called from search order merge processing, instead of merging (merge), in distance order merge processing, clusters included in each grid are merged into pair list (pairList). Processing to add as a candidate cluster (add) is executed. For example, step S717 of the adjacent search process (without upper search) described with reference to FIG. 20 is rewritten as follows in the case of the distance order merge process.

  “If it is determined in step S715 that the distance d is equal to or smaller than the threshold th, the merging unit 107 is included in the cluster included in the element of index i in the grid list (glist) and included in the element of index i + 1. Add the cluster as a merge candidate cluster to the pair list (pairList) (add). "

  Subsequently, the merging unit 107 sorts the pair list (pairList) into cluster pairs as elements in the order of the distance d between the clusters (step S1003). For this reason, in step S1001, in addition to the information on the clusters to be paired, the information on the distance d between the clusters may be added together to the pair list (pairList).

  Subsequently, in step S1005, the merge unit 107 repeats step S1007 in order from the top of the pair list (pairList) according to the order sorted in step S1003 (pair list loop). Here, the index of the element of the pair list (pairList) that is the processing target in the loop of step S1005 is k. The pair list is sorted in the order of the distance d between the clusters, and processing for merging is executed in accordance with the sorted order, so that the merge candidate clusters having the smallest distance are merged in order.

  In step S1007, a new cluster is generated by merging the clusters of the cluster pair that is an element of the index k of the pair list (pairList) (merge). This new cluster information may include information of the original cluster that has been merged. In step S1007, if one cluster to be processed has been merged with another cluster in the loop processing of step S1005 previously performed, the merge is performed with the other cluster to be processed and the merged cluster. Can be executed between. In step S1007, if both the clusters to be processed have been merged with other clusters in the loop processing in step S1005 previously performed, the merging unit 107 does not perform merging, and the pair list (pairList ) May be moved to the next element (step S1005).

  When distance order merging is enabled, a storage area for holding a pair list (pairList) is consumed. The size of the pair list (pairList) is the number of merge candidate clusters that may be merged. Therefore, the size of the pair list (pairList) coincides with the maximum distance calculation count in the worst case (the case where the size becomes the largest), and becomes larger as the range of the adjacent determination becomes wider. Here, the maximum number of distance calculations increases as the number of grids to be processed, the direction of neighborhood search, and the number of higher levels for executing higher level searches increase. Therefore, in an environment where the usable storage area is limited, for example, the merge setting information described with reference to FIG. 13 is used, and the distance order merge is enabled when the number of grids to be processed is less than a certain level. It is desirable to do.

  Heretofore, the merging process in the first embodiment of the present invention has been described. In a general merge process, a distance calculation process occurs for each combination of the entire cluster, and the processing load is high. In general merge processing, it is difficult to adjust the processing load. On the other hand, in the merge processing according to the present embodiment, for example, according to the number of grids 1031 including the cluster 1021, whether to merge, whether search order merge or distance order merge, full match merge or neighborhood search merge. It is possible to select the merge setting and adjust the accuracy and processing load of the merge. In the neighbor search merge, the number of distance calculations is reduced and the processing load is reduced by limiting the clusters that calculate the distance between clusters for merging to adjacent clusters when sorting in a certain direction.

<2. Second Embodiment>
In the second embodiment of the present invention, a three-dimensional space including the earth corresponds to a feature space. In the present embodiment, the position information in the three-dimensional space is expressed using a three-dimensional orthogonal coordinate system such as x-coordinate, y-coordinate, and z-coordinate. In the present embodiment, the cluster is an area including the position information of the content included in the block defined in the three-dimensional space by the x coordinate, the y coordinate, and the z coordinate.

  Note that the second embodiment of the present invention is different from the first embodiment of the present invention in that clustering is performed based on blocks defined by three-dimensional coordinates rather than grids defined by two-dimensional coordinates. However, since the other configuration is substantially the same as that of the first embodiment, detailed description thereof is omitted here.

(2-1. Overview of block-based location clustering)
With reference to FIGS. 30A to 34, an outline of clustering in the second exemplary embodiment of the present invention will be described. Clustering in the present embodiment classifies content having position information into clusters on the basis of blocks defined using a three-dimensional orthogonal coordinate system, and can be said to be block-based position clustering.

(About blocks)
FIG. 30A is a diagram illustrating an example of a relationship among content, clusters, and blocks according to the second embodiment of the present invention. FIG. 30A shows a three-dimensional space 2001, content 2011, cluster 2021, and block 2031.

  A three-dimensional space 2001 is a space including part or all of the earth. In the present embodiment, the three-dimensional space 2001 is a three-dimensional space in which position information is represented by three-dimensional coordinates such as an x coordinate, a y coordinate, and a z coordinate.

  The content 2011 is data having position information for specifying a position in the three-dimensional space 2001. The content 2011 may be position information itself, or may be data to which position information is added as additional information for some information. The content 2011 can be, for example, image data to which position information on the shooting location is added.

  The cluster 2021 is an area including the contents 2011 that are close to each other in the three-dimensional space 2001. The cluster 2021 is illustrated as a rectangular parallelepiped, but may have other shapes. The cluster 2021 may be a solid that circumscribes the content 2011. In the illustrated example, the content 2011 is located on the ground surface. For this reason, the content 2011 is on the cut surface 2021s of the ground surface of the cluster 2021.

  A block 2031 is a block set in the three-dimensional space 2001. The block 2031 may be a rectangular parallelepiped region defined by the x-coordinate, y-coordinate, and z-coordinate ranges in the three-dimensional space 2001. The size of the block 2031 can be set to an appropriate size according to the clustering conditions such as the number of contents 2011 and the size of the area to be clustered.

  As shown in the figure, in the present embodiment, the contents 2011 included in the same block 2031 are classified into the same cluster 2021. That is, in the block-based position clustering process in the present embodiment, whether or not they are included in the same block 2031 is a basic criterion for clustering. The illustrated contents 2011 are all included in the same block 2031 and are therefore classified into the same cluster 2021.

  In the illustrated example, the contents 2011 are all located on the ground surface. However, the content 2011 may be located, for example, inside the earth = underground. In this case, the content 2011 is located behind the cut surface 2021 s of the cluster 2021 due to the ground surface. Further, the content 2011 may be located outside the earth = in the air. In this case, the content 2011 is on the near side of the cut surface 2021s of the ground surface of the cluster 2021. Since the clustering standard is the block 2031 that extends between the inside and outside of the earth, the cluster 2021 can be set for the content 2011 even in such a case.

  FIG. 30B is a diagram showing an example of content and cluster display according to the second embodiment of the present invention. FIG. 30B shows an example in which the content 2011 and the cluster 2021 in the example of FIG. 30A are displayed as graphics on the ground surface.

  In the illustrated example, the content 2011 and the cluster display area 2021d that is a circle circumscribing the cut surface 2021s by the ground surface of the cluster 2021 are displayed. As described above, the cluster display area 2021d can be a circumscribed figure of the cut surface 2021s by the ground surface of the cluster 2021. As described above, when the content 2011 is located underground or in the air, the cluster display area 2021d may be a three-dimensional figure.

  In the block-based position clustering in this embodiment, the position information itself of the content 2011 represents the block 2031 including the content 2011. In the same manner as in the first embodiment described above, when the contents 2011 are sorted in the order of N-ary values in which the x-coordinate, y-coordinate, and z-coordinate values are sequentially arranged one by one, they are included in the same block 2031 area. The contents 2011 are adjacent to each other in the sorting result. Whether or not the adjacent contents 2011 are included in the same block 2031 in the sorting result is, for example, whether the upper k digits (k = 1, 2,...) Of the N-ary value are common. Can be determined. Here, as described above, the contents 2011 included in the same block 2031 are contents classified into the same cluster 2021. Therefore, in the block-based position clustering in the present embodiment, sorting the numerical values generated from the position information of the content 2011 is a main process of clustering.

(About the block hierarchy)
Similarly to the grid 1031 in the first embodiment, the block 2031 in the present embodiment can also have a hierarchical structure. Here, a case where the center coincides with the center of the earth and the block including the entire earth is a level 0 block will be described as an example. In this case, the level 1 block is a block obtained by dividing the level 0 block into two for the x coordinate, the y coordinate, and the z coordinate. Further, the level 2 block is a block obtained by dividing the level 1 block into two for the x coordinate, the y coordinate, and the z coordinate. Further, similarly, the level 2 block area is divided into two for the x coordinate, y coordinate, and z coordinate, and the level 3 block area is divided into two for the x coordinate, y coordinate, and z coordinate. Lower level blocks can be defined, such as level 4 blocks.

  FIG. 31 is a diagram for explaining the division of the ground surface by the blocks in the second embodiment of the present invention. If the center of the level 0 block coincides with the center of the earth, the ground surface 2010 is divided into eight regions 2032a-2032h by the level 1 block as shown. FIG. 31 is a view of the earth as seen from the outside, and 2032a to 2032e are shown in the region divided by the level 1 block.

  FIG. 32 is a diagram for explaining division of the ground surface by blocks according to the second embodiment of the present invention. FIG. 32 is a diagram in which the earth is developed in a plane. The eight regions 2032a to 2032h of the ground surface 2010 divided by the level 1 block and the ground surface 2010 divided by the level 2 block obtained by further dividing the level 1 block. The region 2033 is shown. In the present embodiment, the level 1 block is divided into eight by the level 2 block, but one of the level 2 blocks is inside the earth and thus does not intersect the ground surface 2010. Therefore, the eight regions 2032a to 2032h of the ground surface 2010 divided by the level 1 block are respectively divided by the level 2 block to become 8-1 = 7 regions 2033.

  FIG. 33 is a diagram for explaining the division of the ground surface by blocks in the second embodiment of the present invention. In FIG. 33, the area | region of the ground surface 2010 divided | segmented by the level 5 block among the low-order blocks sequentially defined similarly to said level 0 block-level 2 block is shown in figure. As in the illustrated example, the level 5 block has a size suitable for classifying the content 2011 for each region of Japan, for example. Also in the block-based position clustering in the present embodiment, it is possible to balance the clustering granularity and the processing load by adjusting the block level used for the clustering process.

  As described above, in the present embodiment, a block obtained by dividing a block at a certain level into two with respect to the x coordinate, the y coordinate, and the z coordinate is a block at the next lower level. In other words, the lower level block divides the area of the upper level block into eight. Therefore, the hierarchical structure of the block 2031 in the present embodiment has an octree structure in which a level 0 block is a root node and blocks of the following levels are nodes, and a cluster 2021 defined according to the block 2031 is also: It can be said that it has the same octree structure.

  Here, in general distance-based position clustering, when a tree structure of a cluster is defined, a storage area that holds information on the tree structure is consumed. On the other hand, in the block-based position clustering in the present embodiment, since the octree structure of the block 2031 is uniquely determined as described above, if information on which level the block 2031 is is retained. Based on the octree structure of the block 2031, the tree structure of the cluster 2021 can be easily grasped.

(Comparison with grid base)
In the grid-based position clustering in the first embodiment described above, the ground surface is handled as a two-dimensional plane in which position information is represented by two-dimensional coordinates such as latitude and longitude. In this case, the grid size is defined as, for example, 1 degree latitude × 1 degree longitude. However, as is well known, the distance per degree of latitude is approximately 111 km and is almost constant, whereas the distance per degree of longitude becomes shorter as the latitude becomes higher. Specifically, the distance per 1 degree longitude on the equator at 0 degrees latitude is about 111 km, while the distance per 1 degree longitude at 60 degrees latitude is about 55.7 km. Therefore, when the grid size is expressed by an actual distance, it is 111 km × 111 km near the equator, whereas it is 111 km × 55.7 km near the latitude 60 degrees, which is about half of the area.

  On the other hand, in the block-based position clustering in this embodiment, the cluster 2021 is set by surrounding the content 2011 on the ground surface 2010 with the block 2031 defined in the three-dimensional space 2001. Therefore, the size of the block 2031 does not change depending on the latitude. Note that the size of the area occupied by the ground surface 2010 may be different between the blocks 2031 at the same latitude, depending on the way the blocks 2031 intersect with the ground surface 2010. By performing merge processing together with such block-based position clustering, clustering with a uniform granularity can be realized over the entire earth regardless of latitude.

(N-ary value associated with the block)
FIG. 34 is a diagram for explaining clustering in the second embodiment of the present invention. The number given to each block 2031 is an index when these blocks are sorted by N-ary numerical values.

  Also in this embodiment, an N-ary value is generated by the N-ary value generation unit 101 of the information processing apparatus 100. The N-ary value in the present embodiment is the value of the coordinate of each dimension expressed by a binary number of a predetermined number of digits for data having position information represented by three-dimensional coordinates in the three-dimensional space 2001 for each dimension. It may be a binary value arranged one digit at a time.

For example, when 21 digits are set as the predetermined number of digits, the N-ary value generation unit 101 represents the values of the x-coordinate, y-coordinate, and z-coordinate as 21-digit binary values. Here, the x coordinate expressed in binary number is “x ₂₀ x ₁₉ x ₁₈ ... X ₀ ”, and the longitude expressed in binary number is “y ₂₀ y ₁₉ y ₁₈ ... Y ₀ ”. Assuming that the z-coordinate expressed in binary number is “z ₂₀ z ₁₉ z ₁₈ ... Z ₀ ”, a binary value in which these values are arranged one digit at a time in each dimension is “x ₂₀ y ₂₀ z _20”. x ₁₉ y ₁₉ z ₁₉ x ₁₈ y ₁₈ z ₁₈ ... x ₀ y ₀ z ₀ ”is a 63-digit binary value. When the predetermined number of digits is 21 digits, the minimum resolution in the diagonal direction of the block 2031 is equivalent to 11 m. The predetermined number of digits can be set to an appropriate value in consideration of, for example, the required minimum resolution and the size of the data unit handled by the information processing apparatus 100 (for example, 32 bits, 64 bits, etc.).

In the illustrated example, in the three-dimensional space 2001, the coordinates of each dimension are represented by three-digit binary values. Here, the block 2031 is a block defined by the upper two digits of this binary value. The block 2031 divides the three-dimensional space 2001 by (2 ³ ) ² = 64. In the figure, each block 2031 is given an index of 0 to 63. This index corresponds to the sort order when the block 2031 is sorted in the order of binary values that can be taken by the content 2011 that can be included in the block 2031. In this case, the coordinates and binary values generated by the N-ary value generation unit 101 of the content 2011 that can be included in the block 2031 of the indexes 0 to 8 and 55 to 63 are as shown in Table 2 below. In the table, the notations “x”, “y”, and “z” of “00x”, “00y”, and “00z” indicate arbitrary values (0 or 1).

In the present embodiment, the clustering unit 103 classifies a plurality of contents 2011 having the same upper k digits (k = 1, 2,...) Of the binary values generated by the N-ary value generation unit 101 into the same cluster. To do. When k = 3 × m (m = 1, 2,...), the cluster into which the content 2011 having the same upper k digits of the binary value is classified is m layers of 2 ³ = octtree structure cluster. Eyes.

  Here, referring to Table 2, the contents 2011 included in the blocks 2031 of the indexes 0 to 7 all have a common binary value of “000”. Referring to FIG. 34, it can be seen that the block 2031 including these contents 2011 is eight blocks 2031 located in the lower left back of the figure. As shown in the figure, these blocks 2031 are eight blocks 2031 constituting an upper block 2041, which is a block higher by one layer. Therefore, for example, the content 2011 included in the block 2031 of the index 1 and the content 2011 included in the block 2031 of the index 5 are included in the same upper block 2041.

  Also in the present embodiment, as described with reference to FIGS. 30A and 30B, the blocks 2031 are associated with the clusters 2021 in the same manner as the relationship between the grids and clusters in the first embodiment described above. Yes. Therefore, it can be seen from the above example that the contents 2011 having the same upper k digits of binary values are classified into the same cluster 2021.

  Note that other processing related to clustering and merge processing of the present embodiment can be performed in the same manner as in the first embodiment.

<3. Hardware configuration of information processing apparatus according to embodiment of the present invention>
Next, the hardware configuration of the information processing apparatus 100 according to the embodiment of the present invention will be described in detail with reference to FIG. FIG. 35 is a block diagram for explaining a hardware configuration of the information processing apparatus 100 according to the embodiment of the present invention.

  The information processing apparatus 100 mainly includes a CPU 901, a ROM 903, and a RAM 905. The information processing apparatus 100 further includes a host bus 907, a bridge 909, an external bus 911, an interface 913, an input device 915, an output device 917, a storage device 919, a drive 921, and a connection port 923. And a communication device 925.

  The CPU 901 functions as an arithmetic processing unit and a control unit, and controls all or a part of the operation in the information processing apparatus 100 according to various programs recorded in the ROM 903, the RAM 905, the storage device 919, or the removable recording medium 927. The ROM 903 stores programs used by the CPU 901, calculation parameters, and the like. The RAM 905 primarily stores programs used in the execution of the CPU 901, parameters that change as appropriate during the execution, and the like. These are connected to each other by a host bus 907 constituted by an internal bus such as a CPU bus.

  The host bus 907 is connected to an external bus 911 such as a PCI (Peripheral Component Interconnect / Interface) bus via a bridge 909.

  The input device 915 is an operation unit operated by the user, such as a mouse, a keyboard, a touch panel, a button, a switch, and a lever. Further, the input device 915 may be, for example, remote control means (so-called remote control) using infrared rays or other radio waves, or an external connection device such as a mobile phone or a PDA corresponding to the operation of the information processing device 100. 929 may be used. Furthermore, the input device 915 includes an input control circuit that generates an input signal based on information input by a user using the above-described operation means and outputs the input signal to the CPU 901, for example. The user of the information processing apparatus 100 can input various data and instruct processing operations to the information processing apparatus 100 by operating the input device 915.

  The output device 917 is configured by a device capable of visually or audibly notifying acquired information to the user. Examples of such devices include CRT display devices, liquid crystal display devices, plasma display devices, EL display devices and display devices such as lamps, audio output devices such as speakers and headphones, printer devices, mobile phones, and facsimiles. For example, the output device 917 outputs results obtained by various processes performed by the information processing apparatus 100. Specifically, the display device displays the results obtained by various processes performed by the information processing device 100 as text or images. On the other hand, the audio output device converts an audio signal composed of reproduced audio data, acoustic data, and the like into an analog signal and outputs the analog signal.

  The storage device 919 is a data storage device configured as an example of a storage unit of the information processing device 100. The storage device 919 includes, for example, a magnetic storage device such as an HDD (Hard Disk Drive), a semiconductor storage device, an optical storage device, or a magneto-optical storage device. The storage device 919 stores programs executed by the CPU 901, various data, various data acquired from the outside, and the like.

  The drive 921 is a recording medium reader / writer, and is built in or externally attached to the information processing apparatus 100. The drive 921 reads information recorded on a removable recording medium 927 such as a mounted magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, and outputs the information to the RAM 905. In addition, the drive 921 can write a record on a removable recording medium 927 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory. The removable recording medium 927 is, for example, a DVD medium, an HD-DVD medium, a Blu-ray medium, or the like. The removable recording medium 927 may be a compact flash (CF), a flash memory, an SD memory card (Secure Digital memory card), or the like. The removable recording medium 927 may be, for example, an IC card (Integrated Circuit card) on which a non-contact IC chip is mounted, an electronic device, or the like.

  The connection port 923 is a port for directly connecting a device to the information processing apparatus 100. Examples of the connection port 923 include a USB (Universal Serial Bus) port, an IEEE 1394 port, and a SCSI (Small Computer System Interface) port. As another example of the connection port 923, there are an RS-232C port, an optical audio terminal, an HDMI (High-Definition Multimedia Interface) port, and the like. By connecting the external connection device 929 to the connection port 923, the information processing apparatus 100 acquires various data directly from the external connection device 929 or provides various data to the external connection device 929.

  The communication device 925 is a communication interface configured by a communication device or the like for connecting to the communication network 931, for example. The communication device 925 is, for example, a communication card for wired or wireless LAN (Local Area Network), Bluetooth (registered trademark), or WUSB (Wireless USB). The communication device 925 may be a router for optical communication, a router for ADSL (Asymmetric Digital Subscriber Line), or a modem for various communication. The communication device 925 can transmit and receive signals and the like according to a predetermined protocol such as TCP / IP, for example, with the Internet or other communication devices. The communication network 931 connected to the communication device 925 is configured by a wired or wireless network, and may be, for example, the Internet, a home LAN, infrared communication, radio wave communication, satellite communication, or the like. .

  Heretofore, an example of the hardware configuration capable of realizing the function of the information processing apparatus 100 according to the embodiment of the present invention has been shown. Each component described above may be configured using a general-purpose member, or may be configured by hardware specialized for the function of each component. Therefore, it is possible to change the hardware configuration to be used as appropriate according to the technical level at the time of carrying out this embodiment.

<4. Summary>
(Example of configuration and effect of embodiment)
In the embodiment of the present invention described above, a predetermined number of N-ary numbers (N = 2) are used for data having feature space position information defined by D-dimensional coordinates (D = 2, 3,...). , 3,...), An N-ary value generation unit for generating an N-ary value in which the values of the coordinates of each dimension expressed in order for each of the dimensions are arranged, and the upper k digits of the N-ary value ( There is provided an information processing apparatus including a clustering unit that classifies the data having the same k = 1, 2,.

  According to such a configuration, it is possible to replace the clustering process of data having position information with a process of sorting N-ary values associated with each data. The distance calculation processing between position information is replaced by numerical value comparison processing, and the number of processing times itself can be reduced. For example, clustering requires fewer processors and storage area resources, and clustering processing is accelerated. .

The clustering unit, k = D × m (m = 1,2, ···) when it is, the data higher k digits of the N-ary number is common, m layer of ^{N D-ary} tree structure Cluster You may classify into the same cluster visually.

  According to this configuration, the hierarchical structure of the cluster can be easily configured. Furthermore, since it is not necessary to maintain the entire hierarchical structure, resources in the storage area can be saved.

  The clustering unit may include a clustering sorting unit that sorts the data in the order of the N-ary value, and may specify the data classified into the same cluster from a result of the sorting.

  According to such a configuration, data classified into the same cluster can be easily specified based on the sorting result.

  The clustering unit is a cluster that identifies the one cluster based on a first position where the data classified into one cluster appears in the sorting result and the number of the data classified into the one cluster. Specific information may be generated.

  According to such a configuration, since the cluster specifying information does not necessarily include information for individually specifying data classified into clusters, the resources required for generating and storing the cluster specifying information are reduced, and the clustering process is performed. Is faster.

  The information processing apparatus includes: a merge sorting unit that sorts the clusters with respect to a first direction in the feature space based on a result of a first rank determination process based on the D-dimensional coordinates; and the first direction. An adjacency determination unit that determines whether or not the clusters sorted for are adjacent to each other in the first direction, and further includes a merge unit that merges the clusters determined to be adjacent to each other in the first direction You may prepare.

  According to such a configuration, since the merge is performed on clusters that are determined to be adjacent as a result of sorting, fewer resources are required for processing compared to performing the merge on all clusters. Clustering processing including merged cluster merging processing is speeded up.

  The merge sorting unit sorts the clusters in the second direction in the feature space based on the result of the second rank determination process based on the D-dimensional coordinates, and the adjacency determination unit The clusters sorted in the second direction are determined to be adjacent to each other in the second direction, and the merge unit may further merge the clusters determined to be adjacent to each other in the second direction. Good.

  According to such a configuration, since it is determined whether or not the clusters are adjacent to each other in the two directions in the feature space, the clusters located in the vicinity of each other can be more reliably merged.

  The feature space is a ground surface, the D-dimensional coordinates are two-dimensional coordinates composed of latitude and longitude, and the cluster is included in a grid defined on the ground surface by the two-dimensional coordinates. The first rank determination process sorts the grid in the first direction and gives the sorting order of the grid as a rank to the clusters included in each grid. It may be a process.

  According to this configuration, data having position information on the ground surface can be clustered at high speed using the grid defined by the latitude and longitude. Further, the cluster sorting can be speeded up by giving ranks to the clusters by this grid sorting.

  The feature space is a three-dimensional space, the D-dimensional coordinates are three-dimensional coordinates constituting an orthogonal coordinate system, and the cluster is defined in the three-dimensional space by the three-dimensional coordinates. It may be an area including the position information of the data included in the block.

  According to such a configuration, data having position information in a three-dimensional space can be clustered at high speed using blocks defined by an orthogonal coordinate system. In addition, when the ground surface is divided into blocks, data can be clustered by clusters that do not cause distortion due to latitude on the ground surface.

  In the embodiment of the present invention, the process of merging clusters includes a process of calculating a distance between the clusters, and a process of merging the clusters when the calculated distance is equal to or less than a predetermined threshold. Can be included.

  According to such a configuration, the processing required for the merge processing can be further reduced, and the clustering processing including the processing for merging the generated clusters is further accelerated.

  The process of merging the clusters includes a process of calculating a distance between the clusters, a process of storing the cluster as a merge candidate cluster when the calculated distance is equal to or less than a predetermined threshold, and the storage The merge candidate clusters may be merged in order from the merge candidate clusters with the smallest distance between them.

  According to such a configuration, clusters closer to each other can be reliably merged, and the accuracy of clustering including the process of merging the generated clusters can be improved.

(Modification of the embodiment)
Here, in the first embodiment, the feature space is the ground surface, and in the second embodiment, the feature space is a three-dimensional space including the earth. However, the present invention is not limited to such an example. For example, the feature space is not a real space, but may be a color space such as RGB or YUV, or may be a higher-order feature amount space for expressing an image feature amount.

In the first embodiment, D = 2 and N = 2, and in the second embodiment, D = 3 and N = 2. However, the present invention is not limited to such an example. For example, D may be 4 or more. In other words, the data may have feature space position information defined by four-dimensional or higher-dimensional coordinates. As an example, when D = 4 and the coordinates of each dimension are expressed by 16-digit binary numbers, the N-ary value generation unit generates 16 × 4 = 64-digit binary values. In addition, for example, the N-ary value generation unit may generate an N-ary number represented by using an arbitrary radix such as an octal number, a decimal number, or a hexadecimal number. As an example, when D = 2 and the coordinates of each dimension are expressed in 29-digit hexadecimal numbers, the N-ary value generation unit generates 29 × 2 = 58-digit hexadecimal values, and the data is 16 ² = 256. It is classified into a branch tree cluster.

  In the first embodiment, the two-dimensional coordinates are latitude and longitude, and in the second embodiment, the three-dimensional coordinates are an orthogonal coordinate system including an x coordinate, a y coordinate, and a z coordinate. However, the present invention is not limited to such an example. The two-dimensional, three-dimensional, and D-dimensional coordinates may be an orthogonal coordinate system, an oblique coordinate system, or a polar coordinate system.

  The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field to which the present invention pertains can come up with various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that these also belong to the technical scope of the present invention.

DESCRIPTION OF SYMBOLS 100 Information processing apparatus 101 N-ary value generation part 103 Clustering part 105 Merge part 107 Sort part 109 Adjacency determination part 119 Storage part 1001 Ground surface 1011, 2011 Content 1021, 2021 Cluster 1031 Grid 2001 Three-dimensional space 2031 Block

Claims (13)

The data having the position information of the feature space defined by the D-dimensional coordinates (D = 2, 3,...) Is expressed by an N-ary number (N = 2, 3,...) Having a predetermined number of digits. An N-ary value generator for generating an N-ary value in which the values of the coordinates of each dimension are arranged one digit at a time for each dimension;
A clustering unit that classifies the data having the same upper k digits (k = 1, 2,...) Of the N-ary value into the same cluster;
An information processing apparatus comprising: The clustering unit, k = D × m (m = 1,2, ···) in the case of, the data higher k digits of the N-ary number is common, m layer of ^{N D-ary} tree structure Cluster Classify into the same cluster by eye,
The information processing apparatus according to claim 1. The clustering unit includes a clustering sorting unit that sorts the data in the order of the N-ary value, and identifies the data classified into the same cluster from the result of the sorting.
The information processing apparatus according to claim 1 or 2. The clustering unit is a cluster that identifies the one cluster based on a first position where the data classified into one cluster appears in the sorting result and the number of the data classified into the one cluster. Generate specific information,
The information processing apparatus according to claim 3. A merging sort unit for sorting the clusters in the first direction in the feature space based on a result of a first rank determination process based on the D-dimensional coordinates;
An adjacency determination unit that determines whether or not clusters sorted in the first direction are adjacent to each other in the first direction;
The information processing apparatus according to claim 1, further comprising: a merge unit that merges clusters determined to be adjacent to each other in the first direction. The merging sorting unit sorts the clusters in the second direction in the feature space based on the result of the second ranking determination process based on the D-dimensional coordinates,
The adjacency determining unit determines whether or not clusters sorted in the second direction are adjacent to each other in the second direction;
The merging unit further merges clusters determined to be adjacent to each other in the second direction;
The information processing apparatus according to claim 5. The feature space is a ground surface;
The D-dimensional coordinates are two-dimensional coordinates composed of latitude and longitude,
The cluster is a region including position information of the data included in a grid defined on the ground surface by the two-dimensional coordinates,
The first ranking determination process is a process of sorting the grid in the first direction and giving the sorting order of the grid as a rank to the clusters included in each grid.
The information processing apparatus according to claim 5. The feature space is a three-dimensional space;
The D-dimensional coordinates are three-dimensional coordinates constituting an orthogonal coordinate system,
The cluster is an area including position information of the data included in a block defined in the three-dimensional space by the three-dimensional coordinates.
The information processing apparatus according to claim 1. The data having position information in the feature space defined by the D-dimensional coordinates (D = 2, 3,...) Is expressed by an N-ary number (N = 2, 3,...) Having a predetermined number of digits. Generating an N-ary value in which the coordinate values of each dimension are arranged one digit at a time for each dimension;
Classifying the data having the same upper k digits (k = 1, 2,...) Of the N-ary values into the same cluster;
An information processing method including: The data having position information in the feature space defined by the D-dimensional coordinates (D = 2, 3,...) Is expressed by an N-ary number (N = 2, 3,...) Having a predetermined number of digits. A process of generating an N-ary value in which the values of the coordinates of each dimension are arranged in order of one digit for each dimension;
A process of classifying the data having the same upper k digits (k = 1, 2,...) Of the N-ary value into the same cluster;
A program that causes a computer to execute. A process of sorting the clusters with respect to a first direction in the feature space based on a result of a first ranking determination process based on the D-dimensional coordinates;
Processing to determine whether clusters sorted in the first direction are adjacent to each other in the first direction;
Merging clusters determined to be adjacent to each other in the first direction;
The program according to claim 10, further causing a computer to execute. The process of merging the clusters is
A process of calculating a distance between the clusters;
A process of merging the clusters when the calculated distance is less than or equal to a predetermined threshold;
The program according to claim 11, including: The process of merging the clusters is
A process of calculating a distance between the clusters;
A process for storing the cluster as a merge candidate cluster when the calculated distance is equal to or less than a predetermined threshold;
Among the stored merge candidate clusters, the process of merging in order from the merge candidate cluster having a smaller distance between each other,
The program according to claim 11, including: